Possible partition assignment bouncing in the WAGED rebalancer

[jiajunwang]

It is noticed that, in the WAGED rebalancer, the partition assignment might be bouncing while the LESS_MOVEMENT configuration is set to a very small number compared with the EVENNESS. One of the possible explanations is the partial rebalance does not take the real change of cluster into consideration while calculating. So if the baseline and the best possible assignments diverge, the partial rebalance might keep changing the assignment of partitions.