Restore Hindsight consolidation worker

[linear-code]

Problem

Hindsight retention succeeds, but observation consolidation has stopped. The omp bank has 19,933+ unconsolidated facts; the latest consolidation has been pending since 2026-07-20 and the preceding operation is stuck processing at 24/14,536.

Diagnosis

• API and PostgreSQL health are green.
• The deployed ai-dev-env Helm values use Hindsight chart 0.8.5 with worker.enabled: false, leaving the single API pod's internal worker responsible for async jobs.
• The API pod has restarted since the stalled operation began.
• Hindsight 0.8.5 only recovers processing tasks owned by the current worker ID, so an operation owned by the previous API process remains orphaned.
• A second consolidation remains pending even while retain tasks complete, so consolidation has no reliable dedicated capacity.
• Hindsight 0.8.6 includes the Kubernetes worker-port fix and newer worker scheduling safeguards.

Scope

Upgrade the chart to 0.8.6, enable a dedicated worker with explicit resources, safely recover the orphaned consolidation operation, and retain automatic consolidation.

Acceptance criteria

• Dedicated Hindsight worker is deployed and healthy.
• The orphaned processing operation no longer blocks consolidation.
• A consolidation operation advances its progress timestamp and processed count.
• pending_consolidation decreases from the pre-repair baseline.
• Retain, recall, and API health continue to work.
• Deployment evidence and recovery procedure are recorded.