Rebalance should maintain server group awareness even when they are unequal

Server groups can become unequal after a node failure. 

In many scenarios such a node maybe taken out of the cluster and rebalance run.

This results in rebalance being run over unequal server groups. The requirement is that rebalance should still place the replicas and active of each vbucket in different server groups - even though they are not equal. This is not the case today.