Update Auto Review lifecycle docs for durable runs

[shiny-code-bot]

Summary

Update code-rs/core/docs/auto-review.md and related docs after durable Auto Review semantics land.

Scope

• Document durable run lifecycle and store location.
• Clarify runtime vs inactivity vs stale/superseded/obsolete.
• Describe per-turn ledger and lazy details.
• Document fallback worktree/concurrency policy honestly.
• Document cancellation policy and proof metrics.
• Reconcile old Make Auto Review snapshot-aware and preserve superseded results #76 language with the new Plan: durable Auto Review quality pipeline #324 architecture.

Acceptance Criteria

• Docs no longer imply elapsed runtime alone means stale.
• Docs distinguish long-running, inactive, stale, superseded, obsolete, failed, skipped, and cancelled.
• Docs describe how the assistant sees compact review state and how full detail is retrieved.
• Docs explain what the harness, not the LLM, owns.
• Docs reference proof metrics and dogfood validation surfaces.

Relationships

Parent: #324
Depends on: #325, #327, #329, #330
Evidence archive: #76

Finish Line

Auto Review docs describe durable run records, compact ledgers, freshness semantics, fallback/concurrency rules, cancellation policy, and proof metrics in terms future implementation and dogfood sessions can rely on.