feat(llm-docs): wire the on-disk docs bundle into the extension host

[midleman]

Summary

Wires the docs-bundle cache from PR #15242 into the extension host and exposes it to extensions as positron.docs.getLocalDocs(), so the AI assistant can read Positron docs from disk instead of fetching them from the web.

• Declares the positron.docs API surface plus the positronLlmsDocsUrl product field, overridable at runtime with POSITRON_LLMS_DOCS_URL
• Adds IExtHostDocs: node port adapters over http/fs/yauzl, and an always-undefined web-worker variant
• Adds the trigger layer: launch fetch, getLocalDocs() with a 10s bounded wait, and a mid-session ai.enabled flip
• Gates every path on ai.enabled, read live
• Publishes the extension host's startup-finished moment so the launch fetch stays clear of the eager-activation burst

Release Notes

New Features

• N/A

Bug Fixes

• N/A

Validation Steps

Manually validated against a local fixture server (POSITRON_LLMS_DOCS_URL=http://127.0.0.1:8099) with a throwaway extension calling getLocalDocs():

Scenario	Result
Dev build resolves latest-by-policy	Bundle installed to <userdata>/User/positron-llm-docs/<version>/
ai.enabled: false	No request at all, undefined, no cache directory
ai.enabled flipped true mid-session	Fetch fires with no reload
Two concurrent getLocalDocs() calls	One download, both callers served
Bundle 404	undefined, no cache, no state written
Checksum file 404	Zip fetched, install refused, undefined
Digest mismatch	Install refused, undefined
Release build, exact bundle published	resolution: exact, isExactMatch: true
Release build, exact not yet published	HEAD 404 then latest, resolution: fallback
Relaunch on an exact cache	Zero network requests
Relaunch on a fallback cache	Re-HEADs exact every launch
Exact publishes later	Converges to exact, superseded directory deleted
Unchanged bundle, stored ETag	Single conditional GET, 304, cache kept
CDN 404 with a warm cache	Cached bundle still served
Server slower than the bounded wait	undefined at 10s, download continues, next call served
Workbench profile and server-side cache	Not run: needs a Workbench container

Decisions are logged under the [llm-docs] in Output > Extension Host:

[llm-docs] fetching http://.../positron-llms-latest.zip (latest-by-policy)
[llm-docs] installed 9999.01.0-1 from http://.../positron-llms-latest.zip
[llm-docs] ai.enabled is off; not serving local docs

QA Notes

Depends on #15242 and must not merge before it.

[github-actions]

E2E Tests 🚀
This PR will run tests tagged with: @:critical @:assistant

Why these tags?

Tag	Source
@:critical	Always runs (required)
@:assistant	Changed files

More on automatic tags from changed files.

^{readme  valid tags}