With L1 disabled a new node join may result some entries are not invalidated for non-owner in DIST mode

I have a test case that have 30,000 entries in cache in dist mode with number of owner set to 2 and L1 is disabled. A new node join the cluster, then the total number of entries in all cache is larger than 60,000. Even worse, after I execute cache.clear() on all caches in the cluster, there are still lots of entries left.

The attached unit test can reproduce this issue.