Geo: Gracefully handle replication when there is a primary database node failover

Problem to solve

When running Geo HA with a database cluster in the primary, replication breaks when there is a database failover in the primary cluster. Replication needs to be re-initialized using the new master database node in the primary cluster.

Proposal

Gracefully handle primary database failovers so replication does not break and does not need to be reinitialized.

Links / references

Customer ticket: https://gitlab.zendesk.com/agent/tickets/118204 (internal use)

Relates to https://gitlab.com/gitlab-org/gitlab-ee/issues/10405