Try to get rid of blocking synchronization on cache entry for cache.localPeek operation

Currently all read/write operations for cache entry value use 'synchronized(this)' on cache entry. This can cause very bad concurrency in scenarios when many threads try to read the same entry, or entry read operation suould wait for some long-running update (e.g. EntryProcessor in atomic cache is executed while entry lock is held).

See also https://issues.apache.org/jira/browse/IGNITE-4932