v0.1.1 - API hardening and fallback reliability

Summary

This release consolidates security hardening, deterministic fallback reliability, scoring/config consistency, and documentation alignment completed in issues #7-#12.

Highlights

• API input hardening for /analyze
  • eml_path rejected in API JSON mode
  • structured attachment schema enforced (name/filename)
  • path-like attachment values rejected
• Default API evidence sanitization
  • sensitive evidence details sanitized by default
  • explicit debug_evidence=true for full internal detail
• End-to-end deterministic fallback improvements
  • fallback now covers parse/evidence/router/judge failures
  • fallback_reason added for observability
• Precheck tuning knobs now have effective score impact
• Policy-driven threshold semantics (removed hardcoded phishing floor assumptions)
• Runtime capability metadata now reflects actual SDK availability/remote call capability
• README/manual/design docs aligned with new API boundaries and fallback/evidence contracts

Included work

• #7, #8, #9, #10, #11, #12

Key commits

• 52aa093 - Harden /analyze input validation for Issue #7
• a52f36f - Sanitize default API evidence response for Issue #8
• 44dc217 - Guarantee fallback with reason on parse/evidence/judge failures (Issue #9)
• 8421ba3 - Wire precheck tuning knobs into effective scoring (Issue #10)
• d339beb - Align threshold semantics and runtime capability metadata (Issue #11)
• 8e3459f - Document API input boundaries and fallback/evidence contracts (Issue #12)
• 45e5c61 - Add changelog summary for completed issues #7-#12

Validation

• ruff check src tests docs scripts
• pytest -k 'not hf_phishing_email_balanced_sample'