F1=58.5%: Return-to-open-threads, synthesize grep results, test→prod tracing

Trace-informed edits targeting top 2 FN root causes (42% of FNs):
- Add Step 2.5: Return to open investigation threads after each finding
- Add grep synthesis: explicitly compare related search results
- Add test→production tracing: when finding test bug, check production code
- Relax missing-definition rule for grep-confirmed absences (0 results = evidence)
- Update PLAN.md with full score history through iter 7