Eval review v1: checklist, review helper, 24 tasks reviewed; G0002/G0022 tests fixed; I/E generator definitions

- docs/eval-inceleme.md checklist; harness/review.py
- 19 accept, 1 fix pending (G0019), 4 flagged (I/E tasks solvable without legacy code)
- generator: categories I and E fix/refactor the seed object in place

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014aUaQeLnwbb1zTpN7kHeat

tasks_gen/eval/G0016/review.json

@@ -0,0 +1,9 @@
+{
+ "decision": "flag",
+ "note": "E is weak: new program with complete rules; legacy need not be read. Generator prompt for E fixed (in-place refactor, spec without rules). Options: keep as B-like report task, or regenerate as E.",
+ "by": "Claude",
+ "date": "2026-10-03",
+ "failed_checks": {
+  "6": "zayif"
+ }
+}