We monitor the health of DB replicas, and on a fatal error,
take them out of the pool.
On a separate goroutine, we keep pinging the unhealthy replicas,
and on getting a good response back, we add them back to the pool.
https://mattermost.atlassian.net/browse/MM-50427
```release-note
Mattermost is now resilient against DB replica outages and will
dynamically choose a replica if it's alive.
Also added a config parameter ReplicaMonitorIntervalSeconds
whose default value is 5. This controls how frequently unhealthy
replicas will be monitored for liveness check.
```
Co-authored-by: Mattermost Build <build@mattermost.com>
