fix(review_parser): classify findings into multiple categories (fixes #485)

[mickume]

Summary

Replaces the single-category _detect_security_category() function with _classify_category(), a multi-category keyword classifier that covers security, correctness, compatibility, testing, and configuration. All non-security findings that match a keyword now receive a non-NULL category, reducing the NULL rate from 91% to a small residual for genuinely uncategorised findings.

Closes #485

Changes

File	Change
agent_fox/session/review_parser.py	Add four keyword sets (_CORRECTNESS_KEYWORDS, _COMPATIBILITY_KEYWORDS, _TESTING_KEYWORDS, _CONFIGURATION_KEYWORDS), replace _detect_security_category with _classify_category (priority-ordered multi-category check), keep old name as backward-compat alias, update call site in parse_review_findings()
tests/unit/session/test_review_parser.py	Add TestClassifyCategory (8 tests) and TestCategoryPopulatedInFindings (6 tests)

Tests

• TestClassifyCategory — unit tests for _classify_category() covering each category keyword, the no-match case, priority ordering, and case-insensitivity
• TestCategoryPopulatedInFindings — integration tests verifying parse_review_output() propagates the correct category for each category type and None for unclassified findings

Verification

• All existing tests pass: ✅
• New tests pass: ✅ (14 added)
• Linter / formatter: ✅
• No regressions: ✅ (5009 total, up from 4995)

---

Auto-generated by af-fix.