time of check/time of use race in data path change code of ns_server may lead to deletion of all buckets after adding node to cluster

SUBJ.

In code that changes data path we first check if node is provisioned (without preventing provision-ness to be changed after that) and the proceed with change of data path. As part of change of data path we delete buckets.

So if node gets added to cluster after check but before data path is actually changed, we'll delete all buckets of cluster.

As improbable as it may seem, it actually occurred in practice. See CBSE-1387.