Server clusterStop only shutsdown user caches

Currently Server#clusterStop only shutsdown the caches returned by EmbeddedCacheManager#getCacheNames. This ignores all internal caches except ___script_cache.

We should shutdown the user caches and then all of the internal ones. That way no rebalancing will be attempted on the internal caches if the exit statuses sent to the individual cluster members are staggered due to network delays.