test: finalize 0.5.2 fixture gate and smoke regression polish

[rlippmann]

Summary

• run fixture drift check in CI using the existing fixtures:check path
• pin Python fixture source to v0.6.0 (matches the TS 0.5.2 fixture corpus and avoids unpinned HEAD drift)
• add targeted error-path regression coverage for clarify/idempotent behavior
• tighten smoke assertions by requiring clean stderr and document forced mock mode for deterministic/offline CI execution
• annotate workflow inline to explain why fixture source is pinned to v0.6.0

Validation

• npm test
• npm run build
• FIXTURES_SOURCE=/tmp/context-compiler-v060-local/tests/fixtures/v2 npm run fixtures:check

Notes

• local npm run fixtures:check against an unpinned ../context-compiler checkout may differ if that checkout is ahead of the pinned fixture baseline