[Feature] Add configurable connect_timeout to pgjsonb returner

[co-cy]

Is your feature request related to a problem? Please describe.

salt/returners/pgjsonb.py calls psycopg2.connect() without connect_timeout. When the database is reachable on the network but does not respond — stalled standby, firewall drop, HAProxy or Sentinel mid-failover — the connect call blocks for tens of seconds on the system TCP timeout. Both event_return and clean_old_jobs run inside the master event loop, so one stalled connect cascades into delayed events and slow returns across the whole master.

The other PostgreSQL libraries Salt links against (e.g. psycopg2.connect) already accept connect_timeout, but pgjsonb does not expose it as a config option, so operators have no way to bound the wait short of patching Salt.

Describe the solution you'd like

Add an optional returner.pgjsonb.connect_timeout (seconds) that is forwarded to psycopg2.connect(connect_timeout=...) only when set. With no default, the kwarg is omitted entirely, so libpq's default behaviour stays in force for existing deployments — operators opt in.

returner.pgjsonb.connect_timeout: 5

Describe alternatives you've considered

• Setting a non-zero default (e.g. 30 seconds). Rejected: existing deployments with slow PG failover would silently start failing after upgrade. Backwards compatibility wins.
• Hard-coding a timeout. Same problem, no operator control.
• Using statement_timeout instead. Different scope — that bounds query duration, not connect duration. Both might be useful eventually; this PR covers connect only.

Additional context

The accompanying PR also handles the case where the timeout is delivered as a string (pillar / env) by coercing it to int in _get_options, mirroring the existing treatment of port. Three behavioural tests cover: backwards-compat default (kwarg omitted), opt-in path (kwarg forwarded verbatim), and string-to-int coercion.

Affected branches: 3006.x, 3007.x, 3008.x, master.