chore(infra,web): cap container logs and trim noisy server logs

[juandelperal]

Summary

Production disk was filling up because Podman keeps container logs forever by default (no size cap on either the json-file or k8s-file driver). A few server actions also write a handful of console.log lines per call on hot polling paths, which compounds the problem.

Container-side (the main fix)

Three places create long-lived containers and none of them passed log options:

• infra/deploy/ansible/roles/app/templates/compose.yml.j2 — traefik, postgres, docker-socket-proxy, reaper, flows, cloudflared, and the local-dev web had no logging: block. Added an x-logging: &default-logging anchor (json-file, max-size: 10m, max-file: 3) and applied it to every service. Podman aliases json-file to k8s-file, so max-size is honored on Podman and max-file is honored by Docker.
• infra/deploy/ansible/roles/app/tasks/main.yml — the production web container is started outside compose via podman run for blue-green deploys. Added --log-driver k8s-file --log-opt max-size=10m.
• apps/web/src/lib/spawner/docker.ts — every user workspace container is created via dockerode with no HostConfig.LogConfig. With many users this was the largest source of unbounded logs. Added the same 10MB × 3 cap.

Worst-case per-container disk usage is now bounded to ~30MB.

Application-side (chatty loggers)

• apps/web/src/actions/spawner.ts ensureInstanceRunningAction — was emitting 6 console.log lines per call. This action is polled every 2 seconds while a workspace is starting and on every workspace page load. Removed the trace logs; kept console.error for real failures.
• apps/web/src/actions/opencode.ts sendMessageAction — was logging 5 lines per message (request/response excerpts). Removed.
• apps/web/src/lib/opencode/transform.ts — [transformParts] Unknown part type was dumping the full part object on every unknown part. Any new OpenCode part type could log multi-KB blobs on every streamed message. Throttled to one console.warn per kind per process.
• apps/web/src/lib/spawner/core.ts — dropped two verbose info logs about direct-IP healthchecks that fired on every workspace start; kept the warning paths.

Out of scope

Did not touch user workspace (OpenCode) log verbosity — log rotation on those containers is enough to keep disk usage bounded.

Test plan

• pnpm lint — 0 errors (pre-existing warnings only)
• pnpm test --run — 4442 passed, 15 skipped, 0 failed
• YAML anchor &default-logging and 7 *default-logging aliases verified
• After deploy: podman inspect <container> | jq '.[0].HostConfig.LogConfig' shows { Type: 'json-file', Config: { 'max-size': '10m', 'max-file': '3' } }
• du -sh /var/lib/containers/storage/overlay-containers/*/userdata/ctr.log stays bounded after a few days under load

[Inakitajes]

Thanks for the contribution! A few small adjustments are needed and it has to be brought up to date with main, but I'll keep working on it and get it merged as soon as possible.