Postgres reliability/scale hardening (post-cutover)

[JSONbored]

Tracking issue for Postgres reliability/scale hardening following the self-host Postgres cutover (Redis, Qdrant, backups, Grafana, Prometheus already healthy per that cutover). No prior upstream issue covered this; opened to track the sub-goals below since they span multiple PRs.

Sub-goals

• Backup restore/validation drill — pg_restore --list validation + a guarded opt-in scratch restore, documented. Shipped in feat(selfhost): add a backup restore/validation drill (verify-backup.sh) #2010.
• Postgres observability — Prometheus/Grafana panels + alerts for connection count, lock waits, slow/problem queries, DB size/table growth, dead tuples/autovacuum pressure, and backup freshness. Shipped in feat(enrichment): leftover console.log / debugger analyzer #2015
• PgBouncer readiness guidance — when to enable --profile pgbouncer (already shipped as a pooler, selfhost(infra): PgBouncer connection pooler #1200/Storage & infrastructure adapters (OSS-selectable backends) #977) + a safe connectivity smoke command. No default-on cutover.
• Off-host backup planning — docs/config placeholders first; if implemented, opt-in and off by default (S3/R2/B2 export never runs without explicit env vars).

Each sub-goal ships as its own narrow, tested PR.

[linear-code]

JSO-2128