Nodes should explicitly clear arbiter durableOpTimeAndWalltime

If an arbiter sends a durableOpTimeAndWalltime (e.g due to SERVER-56619), that OpTimeAndWallTime can remain in the system indefinitely, passed around in replSetUpdatePosition commands, even after the arbiter has been corrected. This is because replSetUpdatePosition only advances the optime, and a correct arbiter always has a null optime. Thus for arbiters we should explicitly clear the arbiter's durableOpTimeAndWallTime when we get a heartbeat, to avoid "ghost" optimes staying in the system.

We should also LOG_WARNING if we get a non-null durable optime from an arbiter.