mstore: implement rebalancing

(re)balancing of data distribution, includes:

- non-mirrored MStore with one store to mirrored with N stores
- moving data from an old disk (store) to a new disk (store)
- changing bucket count ratio between stores, because one might get full
- maybe have a "previous" map for quicker rebalancing
- does it need locking or can we support parallel ops while rebalancing?

Note: the current MStore implementation supports writing to SOME stores and reading from ALL stores, so it should always find data, even if we are in the process of stopping to use one store.