Configure connect_timeout for Postgres connections

This is a follow-up for gitlab-org/gitlab#238556 (closed).

The problem: When a single PG replica becomes unavailable (network-wise), the application-side load balancing freezes and requests run into request timeouts.

The mitigation: We configure a connect_timeout to limit the amount of time for a connection attempt. This has now been added to omnibus with gitlab-org/omnibus-gitlab!4555 (merged).

The caveat: Incoming requests continue to run into the connect timeout while the replica is unavailable - every once in a while (when we recheck the replica state, about every 60s). The majority of requests doesn't run into this anymore. See gitlab-org/gitlab#238556 (comment 406212047) for a longer background.

This issue: This is an application setting configured through omnibus. This issue is about getting this shipped to gstg/gprd.