Expose pgx pool acquisition metrics

What

Surface pgx's pool statistics (pgxpool.Stat()) as Prometheus metrics: acquire count and duration, empty acquires, canceled acquires, and configured max connections.

Why

Today the only pool telemetry is gitlab_artifact_registry_database_connection_pool_size{state} (active/idle/constructing). That shows pool composition but not pressure: acquisition waits, pool-exhaustion stalls, and canceled acquires are invisible, which is exactly what an operator needs when the service slows down under load with the database itself healthy. The container registry's database dashboard gets this from Go's go_sql_dbstats_*; AR's pgx pool has the same data behind Stat(), unexposed.