Introduce second level memory cache for all types of cache.

[yuriyz]

One idea which me and @yurem discussed is to introduce 2 levels cache:

1. first level memory (bound to single node)
2. current cache implementation, provider with REDIS, MEMCACHED, NATIVE.

Get entry by key:

1. First check in-memory, if it's there return
2. If not in-memory, call second level cache (currently configured CacheProvider, e.g. NATIVE, REDIS, MEMCACHED).

Idea is on first level keep most used cache objects, and with memory speed is maximum, however that first level of cache is not going to be replicated. We should think whether it will not introduce problems in cluster.

[yurem]

We can add TTL to each object type in first level cache
For example 30 seconds for attributes and scopes. This means that each node will evict these objects withing 30 seconds which is ok for this object type.