Fix: emit the node: protocol in the ESM build banner and format-check .js/.mjs (stacked on #426)

[AmaadMartin]

Please ensure you have read the contribution guide before creating a pull request.

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):
   N/A — no public issue tracks this.
2. Or, if no issue exists, describe the change:

Problem: Two related defects in the repo's own tooling.

1. Every ESM artifact this repo publishes opened with an unprefixed Node
   built-in import. core/build.js and integrations/build.js inject
   import {createRequire as topLevelCreateRequire} from 'module'; as an
   esbuild banner on every format: 'esm' target — that is dist/esm and
   dist/web for both @google/adk and @google/adk-integrations, 198 files
   per target. The repo's convention everywhere else is node:-prefixed
   (core/build.js itself writes import {writeFile} from 'node:fs/promises';
   seven lines up), so a userland package named module cannot shadow the
   built-in. Not a crash — Node resolves both spellings to the same built-in —
   but bundlers and non-Node resolvers cannot identify the bare form as a
   built-in.
2. It survived because it is a string inside a template literal and because
   core/build.js had never been linted: the root lint/format scripts
   globbed **/*.ts only, and eslint.config.js gated js/recommended and all
   globals behind files: ["**/*.ts"], so a .js file resolved to 23 rules,
   zero globals, and no no-undef.

Solution: This PR is stacked, and most of it is not newly authored here.
See the collision note below — the honest summary is that two open PRs already
covered a half each, so this branch incorporates them rather than
re-implementing them, and adds the two things neither one does.

• Incorporated by merge, not re-typed — the banner specifier fix
  ('module' → 'node:module' in both build scripts) is
  #430's two commits, merged
  in verbatim so that PR keeps sole authorship of those hunks.

• Base of the stack — the ESLint scope widening
  (files: ['**/*.{js,cjs,mjs,ts}'], lint/lint:fix → eslint ., and
  deletion of the dead duplicate generateContentAsync in the two js_*
  fixtures) is #426, which
  this PR is based on.

• Authored here (the delta reviewers should read):

  • package.json — format and format:check were still **/*.ts, so the
    .js/.mjs files Chore: lint JavaScript files, not just TypeScript #426 brought under ESLint were still never
    format-checked. Widened both to **/*.{ts,js,cjs,mjs}, matching the lint
    scope exactly, and committed the resulting reformat of the three files that
    had drifted: .prettierrc.js (double → single quotes),
    scripts/auto-assignment.mjs ({ github, context } → {github, context},
    one 82-column if wrapped) and .github/scripts/csat.cjs. All cosmetic;
    eslint.config.js needed nothing because Chore: lint JavaScript files, not just TypeScript #426 already formatted it.
  • .prettierignore — +api-reference/. eslint.config.js ignores
    api-reference/** (typedoc output; not gitignored, and it contains
    typedoc's own bundled minified assets/main.js) but .prettierignore did
    not, so after npm run docs:generate a local format:check scanned that
    bundle and format rewrote it. CI was unaffected — docs:check is
    typedoc --emit none and runs after format:check — so this was a
    local-dev-only trap. Verified by planting a file at
    api-reference/typescript/assets/main.js: before the fix Prettier listed
    it as different while ESLint correctly ignored it.
  • eslint.config.js + 5 fixtures — removed 7 eslint-disable @typescript-eslint/no-require-imports suppressions that the scope
    widening had just turned from inert comments into load-bearing ones. Chore: lint JavaScript files, not just TypeScript #426's
    override exempts CommonJS sources but matched '**/*.cjs' only, so the
    fixtures that are CommonJS by their package.json "type" rather than by
    extension were left papering over the rule at each call site. Fixed at the
    root — the override now also matches
    tests/integration/build_setup/js_commonjs/**/*.js and
    tests/integration/app_loader/app_js/**/*.js (verified accurate:
    js_commonjs/package.json declares "type": "commonjs", app_js/package.json
    declares no "type", so sourceType: 'commonjs' is correct for both) —
    and all 7 suppressions are deleted, including a file-scope
    /* eslint-disable */ in app_js/app.js that was hiding five require()
    sites by itself.
  • tests/integration/build_setup/esm_build_banner_test.ts — a regression test
    that asserts on the shipped artifacts instead of on the build script's
    source text. See "why a second banner test" below; this is not redundant
    with Fix: use the node: protocol for the module builtin in the generated ESM build banner #430's test, and the mutation results prove it.

  Net suppression count for this branch: +0 added, −7 removed.

Collision check (performed before writing any code).
gh pr list --repo AmaadMartin/adk-js --state open --limit 1000 returned 451
open PRs; filtering titles/branches for node:, banner, lint, format,
prettier, eslint, build.js, mjs, module surfaced two direct hits, both
OPEN and MERGEABLE:

PR	Branch	Overlap
#430	fix/node-protocol-esm-build-banner	Lands the banner fix, byte-identical
#426	fix/lint-scope-javascript-files	Lands the ESLint scope widening, and goes further (adds the .cjs block)

Rather than open a third competing implementation, this PR stacks on #426 and
merges #430. Merge order matters for whoever lands these: if #430 merges
first the merge commit here becomes empty; if this PR merges first, #430 becomes
empty. Either is fine, but they should not be reviewed as independent
implementations of the same hunk.

Why a second banner test rather than reusing #430's. #430 adds
build_banner_test.ts, which reads core/build.js/integrations/build.js and
asserts the source text contains the node:module string. That pins the
implementation, not the behaviour: it cannot observe whether the banner is
actually emitted. Demonstrated by mutation 2 below — with the banner block left
completely intact but its guard disabled, #430's test stays green at 2/2 while
dist/esm and dist/web ship no banner at all. The new test reads
dist/{esm,web}/index.js and fails. #430's test is kept, not replaced; the two
are complementary and cost 8 ms together.

Deliberate scope decisions (stated explicitly rather than left as silent
omissions):

• .cjs deviates from the approved spec, deliberately. The spec scoped
  .cjs out of the widened globs because bringing it under ESLint would need a
  separate files: ['**/*.cjs'] block with sourceType: 'commonjs' — but Chore: lint JavaScript files, not just TypeScript #426,
  the base of this stack, already added exactly that block, so the spec's stated
  reason for the exclusion no longer exists here. Leaving it out would ship
  files that are linted but never format-checked. The globs therefore include
  cjs; measured cost is one extra reformatted file,
  .github/scripts/csat.cjs.
• lint-staged still globs **/\*.{js,ts}, so the pre-commit hook does not
  format .cjs/.mjs even though CI now checks them. One-line fix, outside the
  spec's enumerated file list; queued as a follow-up (raw-7402ee44).
• The ts_commonjs fixtures keep their no-require-imports suppressions.
  Those 5 live in .ts files, which were already linted before this change, so
  they are pre-existing and not made load-bearing by it. Removing them is a
  different judgement call (a .ts file using require() is not CommonJS by
  construction the way a .cjs file is) and would be unrelated churn here.
• No .ts source file is modified, matching the spec — the remaining bare
  built-in imports in core/src/artifacts/file_artifact_service.ts and
  dev/src/cli/cli.ts belong to other tasks and are untouched here.

Not a breaking change. No public API, export, type, or exports field
changes. dist/cjs is unaffected (the format === 'esm' guard is untouched).
The widened globs are developer-facing only and touch no published artifact.

Testing Plan

Please describe the tests that you ran to verify your changes. This is required for all PRs that are not small documentation or typo fixes.
Unit Tests:
[x] I have added or updated unit tests for my change.
[x] All unit tests pass locally.

New: tests/integration/build_setup/esm_build_banner_test.ts, 6 cases —
core and integrations × (dist/esm first line, dist/web first line,
dist/cjs contains no topLevelCreateRequire). It asserts on build output, so
it inherits the integration project's existing precondition that
npm run build has run; CI runs npm run build before the test step.

$ npx vitest run --project integration \
    tests/integration/build_setup/esm_build_banner_test.ts \
    tests/integration/build_setup/build_banner_test.ts
 Test Files  2 passed (2)
      Tests  8 passed (8)

Existing fixtures affected by #426's dead-method deletion, re-run in full:

$ npx vitest run --project integration \
    tests/integration/build_setup/build_setup_test.ts -t "js_"
 Test Files  1 passed (1)
      Tests  10 passed | 14 skipped (24)     # 298s

Proof the new test can fail. Two mutations, both run against the built tree.

Mutation 1 — revert the fix (sed 's/node:module/module/' on both build
scripts, rebuild). The four positive assertions fail, the two cjs negative
assertions correctly stay green:

 Tests  4 failed | 2 passed (6)

AssertionError: expected 'import {createRequire as topLevelCrea…' to be
                         'import {createRequire as topLevelCrea…'
Expected: "import {createRequire as topLevelCreateRequire} from 'node:module';"
Received: "import {createRequire as topLevelCreateRequire} from 'module';"
 ❯ tests/integration/build_setup/esm_build_banner_test.ts:36:37

Mutation 2 — banner still in the source but never emitted
(if (format === 'esm') → if (false && format === 'esm') in core/build.js
only, rebuild core). This is the case that justifies adding the test at all:

| Test | Result under mutation 2 |
| ---------------------------------------------- | ----------------------------------------- | --------------------------------------------------------------------------- |
| build_banner_test.ts (#430, source-text) | Tests 2 passed (2) — false negative |
| esm_build_banner_test.ts (this PR, artifact) | Tests 2 failed | 4 passed (6)— the twocorecases fail,integrations correctly stays green |

Both mutations were reverted and the tree rebuilt before committing; the
git diff for core/build.js/integrations/build.js against the merged #430
commits is empty.

Proof the suppression removal is safe because of the config change, not in
spite of it. Same discipline applied to the ESLint config edit: narrow the
override back to files: ['**/*.cjs'] with the 7 suppressions already deleted,
and lint fails —

✖ 11 problems (11 errors, 0 warnings)   # all @typescript-eslint/no-require-imports

11, not 7, because the single file-scope /* eslint-disable */ in
app_js/app.js was covering five require() sites on its own. With the
widened override restored, npm run lint exits 0.

The fixtures whose suppressions were deleted are executable, so they were run,
not just linted:

$ npx vitest run --project integration \
    tests/integration/build_setup/build_setup_test.ts -t "js_"
 Tests  10 passed | 14 skipped (24)

$ npx vitest run --project integration tests/integration/app_loader/
 Tests  6 passed (6)      # includes "App entrypoint with app_js"

⚠️ app_loader_test.ts is intermittently failing on this machine, before and
after this change — measured 3/6 runs failing with the app_js fixture
reverted to its unmodified state, and a comparable rate with it modified. The
edit to that fixture is the deletion of one comment line, which is inert at
runtime. This is the known pre-existing app_loader flake already covered by
#235, #247, #256, #260, #499, #506, #521 and #545; it is not re-queued here.

Repo-wide gates on the exact pushed commit:

$ npm run build        # exit 0, all three workspaces
$ npm run lint         # exit 0  (eslint .)
$ npm run format:check # exit 0  "All matched files use Prettier code style!"
                       #         prettier "**/*.{ts,js,cjs,mjs}" --check

Manual End-to-End (E2E) Tests:
Please provide instructions on how to manually test your changes, including any necessary setup or configuration.

npm install
npm run build

head -1 core/dist/esm/index.js
head -1 core/dist/web/index.js
head -1 integrations/dist/esm/index.js
head -1 integrations/dist/web/index.js
# each -> import {createRequire as topLevelCreateRequire} from 'node:module';

head -1 core/dist/cjs/index.js
# -> /**    (license header only; the ESM banner is not injected into cjs)

grep -rn "from 'module'" core/dist integrations/dist dev/dist
# -> no matches

npm run lint
npm run format:check

All verified locally on Node v22.22.2 / npm 9.2.0 (the engines and CI matrix
are unchanged by this PR).

Checklist

[x] I have read the CONTRIBUTING.md document.
[x] I have performed a self-review of my own code.
[x] I have commented my code, particularly in hard-to-understand areas.
[x] I have added tests that prove my fix is effective or that my feature works.
[x] New and existing unit tests pass locally with my changes.