Expose compact per-turn Auto Review ledger

[shiny-code-bot]

Summary

Expose durable Auto Review state to the assistant as a compact quality ledger, integrated with the broader context-source observability work in #92.

Scope

• Implement AutoReviewLedger projection from durable run records.
• Inject only active/recent actionable review summaries.
• Keep ledger bounded and observable.
• Use freshness/activity semantics: runtime describes cost, not staleness.

Acceptance Criteria

• Ledger is omitted when Auto Review is idle and no recent actionable run exists.
• Ledger includes active run status, snapshot, runtime, last activity, freshness, and concise finding digests.
• Ledger has a tested byte/token cap.
• Stale/long-running labels are based on activity/freshness, not runtime alone.
• /context or prompt ledger observability can show the Auto Review ledger contribution.

Relationships

Parent: #324
Depends on: #325, #326
Related: #92

Finish Line

Assistant turns receive a bounded Auto Review ledger only when active or recent actionable review state exists, with freshness based on snapshot/activity evidence and no raw review output in default context.