Add query_wait_timeout to user settings

[levkk]

Change

Add query_wait_timeout to user settings.

Why

Given two (2) or more read replicas, we want to force the client to failover to a different replica if the replica it's currently trying to use is unresponsive.

This is added because PostgreSQL protocol does not support read timeouts: the client will wait forever until the server answers or TCP keep-alive will break the connection.

Why not use the global query_wait_timeout setting? We have two different kind of users, foreground traffic user and background traffic user. The foreground traffic user needs to failover as quickly as is reasonable/possible. The background user can and should wait until replica recovers. The background user tends to run heavier queries as well, so we don't want to move all that expensive traffic to a different read replica: we are prioritizing the foreground traffic.

Why not implement query_timeout as well/instead? The assumption is if we can assign a query to a server, the server is healthy and will eventually return a result.

Tests

Will write tests if proposed change is acceptable.