Expand description
Initial-connect retry policy for PostgresProvider.
Wraps the first PgPool connect in a bounded exponential backoff so
transient startup races (Postgres still booting, SSL handshake racing
the TCP listener) recover without surfacing as user-visible failures.
The retry loop intentionally targets a narrow set of error shapes so
permanent failures (auth, missing database, bad URL) fail fast. The
backoff itself runs on crate::resilience::retry::retry_async.