Automatic switchover if master server is behind slave

A semisync primary configured with rpl_semi_sync_master_wait_point=AFTER_SYNC and init-rpl-role=SLAVE will typically truncate its binary log from non-confirmed (missing replica acknowledgement) transactions when recovering after a crash. This is fine if failover has already happened:
1. Either the new primary has the transaction, and the old primary can download and re-apply them. In this case the client needs to do some digging and confirm that the writes went through.
2. The transactions never reached a replica and are totally lost. This is fine, as the client never received an OK.

If, however, the primary recovers before failover, we have an issue: The server may throw away transactions that a replica (or multiple) already has. This causes diverging replication. Ideally, MaxScale should detect this situation and switchover to the most up-to-date replica.