Tolerate transient database connection failures in Oban peer election

Hive now survives transient database connection failures that previously could terminate the Oban peer process. The built-in Oban database peer has been replaced with a local variant that logs exhausted election transactions and schedules the next election instead of crashing, so a brief Postgres blip no longer kills the replica responsible for Oban leadership. Production database sockets also enable client-side keepalives with a shorter probe cadence on Linux, reducing dead idle socket errors when connections close unexpectedly. Replicas that lose a leadership check temporarily stop acting as leader until the next successful election, avoiding duplicate leaders. No migration or operator action is required.