ns_server rebalance algorithm chooses a non-optimal bucket map during rebalance

scenario : add 8th node to a cluster with 7 node where each node in average had 150 vbuckets.

after rebalance we should have 128 vbuckets in each node

vbucket count in some nodes dropped from 128 to 50 and on the other nodes # of vbuckets increased from 128 to 240

its okay to try to move one or two more vbuckets than we should but forcing the cluster to do 200% more rebalance work isnt very nice.