Address PR #252 review findings: error handling, tests, type design

[neuromechanist]

Context

PR #252 (Release v0.8.0: develop -> main) was reviewed with 5 specialized agents. The review found legitimate issues that should be fixed before merging to main.

Important Issues (fix before merge)

Error Handling

• 1. Missing CorruptMirrorError/ValueError handling in 4 mirror endpoints (mirrors.py:164,205,225,277)
• 2. Cost check fails open for unknown models (community.py:621-628)
• 3. Missing OSError handling in create_mirror_endpoint (mirrors.py:129-137)
• 4. One stuck mirror blocks cleanup of all others (mirror.py:409-417)

Code Quality

• 5. asyncio.get_event_loop() deprecated; use get_running_loop() (mirrors.py:249)
• 6. Missing field_validator on RefreshMirrorRequest.community_ids (mirrors.py:89-94)
• 7. Broad except Exception in _cleanup_mirrors (scheduler.py:302-308)

Test Gaps

• 8. No HTTP-level tests for mirror routing middleware
• 9. No HTTP-level tests for mirrors API router
• 10. No test for active_mirror_context context manager
• 11. No test for MirrorInfo.__post_init__ validation
• 12. No test for run_sync_now with invalid sync_type
• 13. No test for TTL clamping in create_mirror

Comment Fixes

• 14. Fix enable_caching docstring ("enabled" -> "requested")
• 15. Fix asyncio.to_thread comment (code uses run_in_executor)
• 16. Fix get_model docstring example (claude-3-5-sonnet -> claude-3.5-sonnet)

Type Design Improvements

• 17. Make MirrorInfo a frozen dataclass
• 18. Move is_safe_identifier to shared utility
• 19. Add non-negativity check to ModelRate.__new__
• 20. Expand SecureFormatter key patterns for other providers

Origin

PR #252 review findings from code-reviewer, silent-failure-hunter, test-analyzer, comment-analyzer, and type-design-analyzer agents.