fix(native-tools): render scalar prompt examples as CLI values

[aaight]

Summary

Fixes https://trello.com/c/6a006a843c7b373ca6092284.

• Split native-tool prompt example rendering by CLI value shape instead of JSON-stringifying every non-boolean example.
• Keep object and array-of-object examples as shell-quoted JSON payloads, while rendering enum/scalar/number values as CLI values and primitive arrays as repeatable flags.
• Added regression coverage for synthetic manifests, real CreatePRReview guidance, enum grammar across shipped definitions, and JSON-shaped comments.
• Updated the gadget authoring guide and changelog with the enum example pitfall.

Verification

• npx vitest run --project unit-backends tests/unit/backends/shared-nativeToolPrompts.test.ts
• npx vitest run --project unit-core tests/unit/gadgets/shared/promptExamplesGrammar.test.ts
• npm test
• npm run lint:fix
• npm run lint
• npm run typecheck
• Manual: rendered CreatePRReview guidance shows --event APPROVE, create-pr-review --help remains runnable, and raw APPROVE / REQUEST_CHANGES / COMMENT parse through to the runtime path.

🕵️ codex · gpt-5.5 · run details

[codecov]

Codecov Report

❌ Patch coverage is 82.05128% with 7 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/backends/shared/nativeToolPrompts.ts	82.05%	7 Missing ⚠️

📢 Thoughts on this report? Let us know!

[nhopeatall]

LGTM - the shared prompt renderer now formats enum/scalar/number examples as CLI values, preserves JSON payloads for object-shaped flags, and keeps primitive arrays aligned with the repeatable flag parser. Targeted prompt-renderer and grammar regression tests pass locally.

🕵️ codex · gpt-5.5 · run details