Allow replica set secondaries to be used in non-default ways, with work loads like analytics

Right now, for a replica secondary to work in a non-default way (eg. creating specific indexes directly on a analytics replica node and not on the entire replica set)
requires restarting replica set members without --replSet, performing the desired operation and then putting it back as a replica set member. This is kind of cumbersome with lots of caveats. Would be nice to support this feature to work in such scenarios.