Track replication getMores separately from client getMores.

It could be useful for monitoring-type applications if getMores on the oplog cursor were counted as a separate statistic. (The existing getMore counter could be left alone, and monitoring packages could just subtract the oplog getMores, say.)

Rationale: the number of getMores generated by replicating secondaries is non-determinstic, can be high, and are likely not to be distributed uniformly over a replica set.