Primary crash and restart within auto-failover window should not lose durable write

Assume there is one replica and all writes are using replicateTo = 1. Expectation is that writes that have been acknowledged to have been replicated should not be lost if primary crashes.

However that is not the case for following scenario (there is another scenario already tracked by MB-28525):

• Primary node crash and restart within failover window: If the master crashes (losing the in memory write) and then comes back up before the auto-failover window, it will re-establish itself as the master. However, it would have lost the in-memory write and when the replica reconnects it will rollback. A workaround is to also force persistTo=1 but that will hurt performance. Also, one failure should have been tolerated correctly under this case (without requiring explicit persistTo=1).