Rather than have the heartbeat code switch secondaries to the RECOVERING state when all nodes appear down and an auth error is detected in at least one, it would be more to the point for the applier to take this step. The point of switching to recovery is to prevent slave-ok reads from reading data that is growing only more stale due to a misconfiguration, but the place where we know this is happening is the applier, not the heartbeat code.

The new heartbeat code does not do the switch to RECOVERING in this circumstance, so this ticket is required to complete the replication refactor.