[grikomsn]

Summary

A working Local-chat picker does not prove Copilot SDK discovery. Add contract tests for the SDK's exact reasoningEffort schema key, its VS Code 1.141 effort filtering, and translation of selected values into native requests. Test discovery with and without Context Window controls.

Cover graded efforts and native toggle/budget models, plus endpoint-native request translation. Document that off/on/auto and separate budget controls remain Local-chat features.

This carries the SDK compatibility coverage from OpenAI PR #77. The runtime already publishes the expected key, so this PR adds tests and documentation without changing its effort schema or transport behavior.

Include a patch Changeset so the SDK compatibility documentation and coverage ship in the next extension release.

Verification

• Focused Copilot SDK contract tests passed.
• Replacing the compiled schema key with mode made the new discovery test fail; the compiled file was restored afterward.
• npm run check passed through the npm run package prepackage hook: 185 tests, zero failures.
• npm run package and git diff --check passed.
• Validation uses pure contract and request-translation tests against pinned VS Code source; this PR does not claim fresh authenticated SDK inference.

Checklist

• Changes are focused.
• Tests pass.
• Packaging passes.
• Relevant documentation is updated.
• Patch Changeset included for the next compatibility release.