Follow-up ops to the docs cache-bloat fix (#3287 + #3373, both merged; deployed via prod-2.166.6-20260601211523). These need infra access — not a pilot task.

Checklist

• Reclaim the stale host volume (do after the prod-2.166.6-… GitLab deploy-prod is green and the new container — with no cache mount — is running):

  du -sh /data/quantflow/pilot_cache   # confirm size before
  rm -rf /data/quantflow/pilot_cache

  Nothing references it anymore (fix(docs): drop persistent .next/cache volume from prod compose #3373 removed the bind mount from docker-compose.prod.yml).
• GitLab Container Registry retention — prod-* tags + untagged :main layers on gitlab.com/quant-flow/pilot-docs accumulate forever. Add a cleanup policy: Settings → Packages & registries → Clean up image tags → keep N most recent, expire untagged.

Context

Root cause: docker-compose.prod.yml bind-mounted /data/quantflow/pilot_cache → /app/.next/cache; with Next's on-server image optimizer on (default), .next/cache/images grew unbounded → disk full. Both vectors fixed in-repo; this issue covers the one-time host reclaim + ongoing registry hygiene.