Avoid writing to a shared CacheStore during rehash

When a node receives new data as new owner during a rehash, it has to write this data to the underlying cachestore when each node has a different store, but it's actually writing to the store also in case the store is shared.

This brings a small performance penalty, but also makes it possible for values which where originally stored using Flag.SKIP_STORE to be actually written to permanent store during a rehash.