Add KEDA-driven HPA on core-worker (Asynq queue depth)

[tayebmokni]

Summary

Configure KEDA-driven HorizontalPodAutoscaler for core-worker targeting Asynq queue depth as the external metric, instead of CPU. Workers should scale on asynq_queue_depth{queue="default"} (target 50/pod) and {queue="media"} (target 20/pod) — pure CPU autoscaling produces bad behavior for throughput-shaped workloads.

Design reference

• docs/09-deployment-ops.md §4.6 (core-worker HPA on queue depth)
• docs/12-jobs-cron.md §11.3 (queue depth metrics)

Acceptance criteria

• KEDA ScaledObject (or HPA via metrics adapter) targeting core-worker Deployment
• Triggers: gonext_asynq_queue_depth{queue="default"} target 50, {queue="media"} target 20
• minReplicaCount: 3, maxReplicaCount: 15
• terminationGracePeriodSeconds: 300 on worker pods preserved
• No readinessProbe on workers (no Service); livenessProbe only
• Verified scale-out when a queue is artificially backed up
• Documented in values.yaml

Dependencies

#66, #82 (metrics exporter)

Complexity

M

[tayebmokni]

Closing as DONE — implementation landed across the 280+ PRs shipped to date. See triage-report.md (2026-05-25 backlog audit) for the specific file/package evidence. Re-open with a specific gap if any acceptance criterion is missed.