fix(postgres): bump backup deadlines, expand repo1 PVC, tighten repo2 retention

[JoiDaggir]

Context

Apr 21 incident: repo1 Longhorn PVC filled (live 500 Gi vs declared 300 Gi) → WAL archive-push to repo2 MinIO broke → primary pg_wal accumulated. Applied live: retention cuts (~129 GB freed), Tier 1 Longhorn cleanup (~716 GB freed). This PR codifies the sustainable settings so drift doesn't recur.

Changes

• CronJob activeDeadlineSeconds: full 3600 → 7200, diff 1800 → 5400. Healthy runs observed at 28–45 min since the Apr 14 cluster upgrade; the old tight deadlines were silently killing jobs via DeadlineExceeded.
• repo1 PVC: 300 Gi → 1000 Gi (live was 500 Gi via manual expansion; Git was out of sync). Accommodates ~40 GB/day WAL + retention + margin.
• repo2 retention: retention-full 8 → 4, retention-diff 14 → 7, retention-archive 4 → 2. Matches what was applied live via pgbackrest expire. PITR window: 4 weekly restore points + ~14 days continuous.

Verify post-merge

1. ArgoCD syncs within ~3 min.
2. Longhorn online-expands repo1 PVC 500 → 1000 Gi (no downtime, no pod restart).
3. Next scheduled CronJob run completes inside the new deadlines.

Out of scope (separate tracks)

• pgBackRest upgrade — not needed (99.9% pipe_w hang fixed operationally by yesterday's cxs-pg-repo-host-0 pod restart).
• rvfc-0 replica reinit (TL 24 vs 28 divergence from Apr 14 upgrade).
• Grafana metrics backend (dead Prometheus datasource since Rancher Monitoring removal).
• MinIO admin-side cleanup (mc rm --recursive for empty-shell prefixes, mc admin info for capacity visibility) — for Gissur.

🤖 Generated with Claude Code