[Bug] Operator should DECOMMISSION BE when Scale-in the cluster

[dengliu]

Currently, operator just simply shutdown BE node when updating the replica number of BE. This will cause data loss. It should be changed to use DECOMMISSION to scale-in the cluse. As DECOMMISSION is an async operation, operator must monitor its status until it is completed before proceeding with the lower-level Kubernetes statefulset scale-in.

cc @kevincai @shileifu @imay

[milletnis]

We second that! Currently we always have to do manual decommission before we reduce the replica in the statefulset