update benchmark logs

Phase A: Full 320-test baseline — L1/L2/L3 complete (75.6% overall pass rate)

Cleanup pass 3: Delete debug_verdict.py, rewrite tasks/todo.md and src/README.md for v2.0

Cleanup pass 2: Fix .gitignore corruption, archive docs/audit + phase10 helpers, fix PROJECT_DESCRIPTION.md

Phase A: Upgrade L1 with TP/FP/FN/TN verdicts + word edits, add F1 scores to matrix

Phase A: Fix verdict logic - strip terminal punctuation before correction comparison

Phase A: Add v2.0 3-level test harness (L1 raw, L2 solo, L3 integrated + master matrix)