Fix: print usage for a bare adk invocation instead of only the version - #597

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
2. Or, if no issue exists, describe the change:
   Problem: The root commander command has an unconditional action handler that prints the version. Bare adk therefore answers with a version string and never lists its subcommands, so the CLI has no discovery affordance at its own front door. Every generated usage line also reads Usage: ADK CLI web ..., which names no runnable command.

Solution: The handler now branches on the --version flag and falls through to program.outputHelp(). The help goes to stdout and the process exits 0, because a bare adk is a discovery request rather than a usage error. I also renamed the program to adk, so every usage line names the real binary.

Why stdout and exit 0

adk exits 0 with help on stdout, but adk deploy still exits 1 with help on stderr. The asymmetry is deliberate:

• The root must keep an action handler, because that handler serves -v/--version. Printing help and returning normally is the natural expression there, and it exits 0.
• The groups deliberately have no action handler, because adding one switches off commander's unknown-subcommand detection. Their exit-1 behaviour follows from that choice.
• gh (0), npm (1) and git (1) all print the usage block on stdout for a bare invocation. The ecosystem is split on the exit code but not on the stream. Exit 0 is also the less breaking option, since bare adk exits 0 today.

Rejected alternative: stderr and exit 1, for symmetry with the groups. Please push back if you prefer it.

outputHelp() is load-bearing. help() raises a CommanderError through commander's exit handling, and help({error: true}) also redirects to stderr and exits 1.

Notes for the reviewer

• This is not an adk-python parity fix. An earlier revision of this branch claimed it matched click's no_args_is_help default. That claim was wrong and I removed it. The default is real, but click 8.2.0 changed the exit code to 2 and emits the usage on stderr (Help shown via no_args_is_help results in exit code 2, was 0 pallets/click#1489), and adk-python resolves to click 8.2+. The Python CLI's bare invocation is a usage error today, which is the opposite of this contract.
• Behaviour change: anything scraping the version out of bare adk breaks. adk --version is the documented spelling and is unaffected, and the exit code stays 0.
• Known adjacent issue, not fixed here: -v means --version on the root and --verbose on the subcommands, and the root wins, so adk web -v silently swallows the flag. It predates this change and is queued separately.
• Deviation from the plan: the reference diff carried a 7-line rationale comment. I kept the 3 lines that name a real trap (outputHelp vs help) and moved the group-asymmetry rationale here, where the plan also requires it.
• Merge note: two branches touch the same two files. fix/cli-root-help-command adds program.helpCommand(true); fix/cli-group-bare-invocation-contract adds a comment on the deploy group. Both merge cleanly against this except for the commander import line, which all three touch. Whoever lands second resolves it to the union.
• Collision check: I listed all 706 open PRs on the fork and diffed every CLI-adjacent one. Fix: restore 'adk help' and 'adk help <command>' at the CLI root #816, Test: pin the usage-error contract for bare intermediate command groups #705 and Fix: document the service URI schemes, the DATABASE_URL fall-back and the memory:// defaults in adk --help #385 touch these two files but change other things; none implements this fix. Fix: name the unknown subcommand instead of reporting too many arguments (stacked on #597) #706 is stacked on this branch and builds on this action handler, so its author must re-stack after the force-push.

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.

npx vitest run --project unit:dev dev/test/cli/cli_test.ts -> 29/29 pass.

One new describe('bare invocation') block with three cases: bare adk prints the usage and all six subcommands and no version; -v and --version print only the version; an unknown subcommand exits 1 with nothing on stdout. The block sits before describe('command: version') to avoid the insertion point another in-flight branch uses. No existing test, mock or hook was modified or deleted, and the diff on the test file is additions only.

Coverage of the new code is 100% line and branch, read from coverage-final.json: both arms of the one new branch are hit, 4 times (version) and 1 time (bare).

Mutation proof. Every new test fails against mutated source:

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
E=dev/dist/esm/cli_entrypoint.js
node $E 2>/dev/null            # usage block on stdout, exit 0
node $E -v                     # 1.6.0, exit 0
node $E --version              # 1.6.0, exit 0
node $E bogus                  # error on stderr, exit 1, stdout empty
node $E deploy                 # deploy usage on stderr, exit 1 (unchanged)
node $E web --help | head -1   # Usage: adk web [options] [agents_dir]

All six behave as listed. The bare output is byte-identical to --help, and it survives 2>/dev/null, which proves it goes to stdout.

Also run on this commit: npm run lint clean, npm run format:check clean, npx tsc --noEmit reports 0 errors.

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.

CI note

The first windows run failed on tests/integration/app_loader/app_loader_test.ts ("discover apps vs agents"), which timed out against its own 40s budget. That test is a known flake and this branch does not touch it: the diff is two CLI files, ubuntu and macos passed the same commit, and 3138 of 3139 tests passed on windows. Four open PRs already target this flake, including #247 ("stop app_loader discovery test timing out on Windows/macOS CI"). The job passed on re-run, and all three platforms are now green.