Align auto-review durable store and lifecycle with Every Code

[shiny-code-bot]

Finish Line

Codex Lab auto-review durable store and lifecycle model is intentionally aligned with the authoritative Every Code source, with Codex Lab-specific improvements explicitly decided.

Context

Issue #126 paused the low-to-high PR audit at PR #63
(native-detached-auto-review-runtime) because the Codex Lab implementation
appears to use a simplified persistence/lifecycle model instead of Every Code's
real review_store.rs / review_coord.rs model.

PR #128 and PR #129 corrected the scoped store layout and migration/awareness
regressions, but they did not answer the larger PR #63 parity question.

Current Status

Completed. PR #136 landed the final target-proof repair on top of the earlier durable-store and lifecycle slices.

Done across the final slice:

• Added snapshot proof fields to AutoReviewRunTarget with legacy persisted-data compatibility.
• Made durable background duplicate reuse use full target freshness proof instead of only branch/head/fingerprint.
• Preserved the broader branch/head matching path for cleanup and supersession where exact proof would be too narrow.
• Bumped the scoped snapshot epoch only after acquiring the background review coordination lock, then refreshed the full persisted target before preparing/spawning the detached review task.
• Treated epoch 0 as no proof so legacy runs remain compatible until a real epoch bump exists.
• Wired app-server summary/detail reads to collect the scoped snapshot epoch so stale completed runs are not surfaced as current through the browser/app-server path.
• Added regression coverage for legacy no-proof targets, epoch/commit mismatch freshness, proof-aware duplicate rejection, target epoch collection, and updated app-server/core fixtures.

Validation:

• cargo test -p codex-auto-review snapshot
• cargo test -p codex-core collect_target_records_current_snapshot_epoch
• cargo check -p codex-app-server
• cargo fmt --all --check (passes with existing stable-rustfmt warnings)
• git diff --check
• Read-only agents reviewed the target-proof repairs before merge; their final app-server epoch finding was fixed before PR Add auto-review target proof #136 was merged.
• PR Add auto-review target proof #136 CI passed all 5 required checks and merged at 318bfcc5ee8d778970c08df7cc366f8233b0775e.

Remaining follow-up is no longer part of #130: any broader ledger/detail shape work or additional Codex Lab-specific improvements should be planned separately from this completed Every Code alignment repair.

Scope

• Compare Every Code review_store.rs, review_coord.rs, related TUI/agent
  lifecycle, and any review housekeeping behavior.
• Compare Codex Lab codex-auto-review, review_persistence.rs, background
  auto-review session state, app-server surfaces, TUI surfaces, and tests.
• Identify which behavior is required parity, which behavior is a Codex Lab
  adaptation worth keeping, and which behavior needs an explicit user decision.
• Split implementation into small PRs if the alignment work is larger than one
  safe change.

Guardrails

• Authoritative Every Code source is ../code-prealign-new-skills/code-rs.
• Do not use ../code as the feature source for this workstream.
• Run read-only comparison first; discuss decisions before implementing product
  improvements that go beyond parity.
• Keep existing valid Codex Lab improvements where they do not conflict with
  the authoritative model.

Validation Plan

• Focused unit/integration tests around review store schema, lifecycle state,
  duplicate/supersession behavior, stale/lost/orphan handling, detail retrieval,
  awareness injection, and app/TUI surfaces touched by the alignment.
• Read-only agent review before merge for each corrective PR.