[codex] Record decision trace replay audit

[YSCJRH]

Summary

• Record the post-Phase-67 decision-trace replay audit as an audit/test closeout, not a Phase 68 opening.
• Add runner-level regression coverage that reuses an existing decision_trace.jsonl and asserts appended replay rows use replay_cache / accepted_from_replay with stable selected indexes and input hashes.
• Tighten README copy from “every decision” to “every materialized generated-run decision” to avoid overclaiming session root nodes before generation.

Boundary

This keeps the queue paused. It does not change decision_trace.jsonl shape, scenario DSL, claim labels, report claim evidence_ids, run trace shape, compare artifact shape, session/node manifest shape, public demo artifact layout, plugin MCP contract, route ownership, or artifact layout.

Validation

• python -m pytest backend\tests\test_post_phase67_decision_trace_replay_audit.py backend\tests\test_post_phase67_runtime_created_world_eval_proof.py backend\tests\test_post_phase67_outcome_eval_audit.py backend\tests\test_post_phase67_successor_intake_audit.py backend\tests\test_decision_kernel.py backend\tests\test_pipeline.py -q -> 41 passed
• python scripts\check_no_secrets.py -> Secret scan passed
• python -m backend.app.cli audit-github-queue --repo YSCJRH/mirror-sim -> paused, no active milestone, failures []
• git diff --check -> exit 0, CRLF warnings only

Review

• Context explorer: no source-backed Phase 68 blocker; recommended audit/test closeout only.
• Evidence auditor: no implementation queue; keep eval-ownership as TODO[verify].
• Final read-only reviewer: no blocking findings after diff review.