Refactoring of Cellar internals: use level triggering instead of edge triggering

Generally speaking, the Cellar internals and logic are way too much complex.
The main drawbacks are:
1. The cluster events (via the Hazelcast topic) contains all update data
2. The distributed maps are not really used (just to get the state)
3. Lot of Cellar components are useless from an user perspective (producers, handlers, ...)

Instead, it would make more sense:
1. To store all in distributed maps
2. To have kind of distributed map listeners on node to "react" when a change is performed (instead of using the topic)
3. Remove the command/topic usage

I will start a PoC around that.