Reclaim abandoned workspace patch staging artifacts durably

[flyingrobots]

A process exit after staging and before rename can retain the deterministic sibling staging artifact introduced by #711.

The existing reconciler must remain read-only and effect-free; deleting during reconciliation would create an unrecorded mutation. Add a separately requested, durably admitted cleanup operation or equivalent scheduler-owned effect whose exact target is derived from the settled attempt ID and whose outcome is durably settled before resumption.

Acceptance:

• no directory sweep or ambient path authority;
• request-before-cleanup and settlement-before-resumption;
• exact attempt-derived staging aperture;
• recovery and idempotent reconciliation;
• replay does not repeat cleanup;
• crash-boundary tests cover the retained staging artifact.

Follow-up from CodeRabbit thread PRRT_kwDOQH8Wr86VD_gm on #712. Not a blocker for #711 because the requested mutation does not belong in the read-only reconciler.

[coderabbitai]

🔗 Related PRs

#688 - Route executable-operation Actions through scheduler-owned Ticks [merged]
#708 - Reconcile external-action settlement retries [merged]
#710 - Admit observation uncertainty without workspace authority [open]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!