Implement pg-boss completed jobs cleanup to prevent table bloat

[ygrishajev]

Problem

The pgboss.job table in the notifications service grows unbounded with completed jobs. In production we observed ~791K rows (440K created, 350K completed), causing:

1. 24MB temp files on every poll query — the pg-boss work query (SELECT ... FOR UPDATE SKIP LOCKED) cannot sort in memory, creating ~24MB temp files every ~250ms
2. Connection drops — excessive I/O from the bloated table causes Broken pipe / connection to client lost errors
3. Container restart loops — connection loss triggers pg-boss stopped event, health check fails, orchestrator kills container, backlog grows

The maintenance cycle that should clean up completed jobs never completes because pg-boss keeps crashing and restarting before it runs.

Action needed

Implement proper cleanup of completed jobs from the pgboss.job table.