Improve replicated adapter to ensure better consistency across the nodes

[cabol]

Improvements

• When a new node joined, it is required to fetch the cached data from any of the cluster nodes. Since in a replicated topology all nodes should have the same data, it is important to ensure all write operations are stopped during the synchronization process avoiding possible inconsistencies. To do so, set a global lock across the cluster when a replicated cache node is joined and the synchronization starts, and then delete it when it ends.
• When using a generational cache as the primary store, ensure the data is the same across all cluster nodes during the synchronization process.

[cabol]

This also should help to have better consistency for cache stats, which was mentioned in issue #77