Implement native Auto Review artifact storage #55

Current Status

Completed by PR #60, merged as 3dee2dd27a8621e4618edda81101fcefcc80a206.

The native Auto Review artifact storage contract now lives in codex-auto-review as a small standalone Codex crate. The runtime trigger, app-server/TUI surface, and Code Bridge work remain intentionally separate.

Finish Line

Codex Lab has native Auto Review run and finding storage with bounded summary/detail semantics, freshness classification, stable IDs, and no dependency on removed plugin state.

Acceptance Criteria

• Define native Auto Review run metadata: run ID, status, source, branch, head/base commit, snapshot/worktree identity when applicable, target, model, timestamps, error summary, and freshness/currentness classification.
• Define native finding digest/detail records with stable finding IDs, bounded summary rows, omitted-count markers, and detail lookup that enforces byte caps.
• Preserve Define Auto Review proof loop MVP #35 semantics: current findings surface, stale/detached/clean runs stay quiet by default, and truncation is explicit when byte budgets cut off output.
• Store under Codex-owned state rooted in the active product home, not under .codex/plugins/data or other removed plugin paths.
• Add focused tests for safe IDs, byte caps, omitted counts, current/stale/detached classification, restart/load behavior, and malformed/missing detail handling.
• Avoid model-context injection in this slice. If a later slice injects findings, it must use bounded core/context fragments.

Relationships

Child of #53. Followed by native detached Auto Review orchestration and app-server/TUI surfacing. Related to #35 and #54.