Persistent implementation for partition store

Raft supports some kind of rebalancing out-of-the-box by sending pieces of its log from up-to-date nodes to a fell behind nodes.

But when there is not enough Raft log available (because Raft took a snapshot in storage and compacted its log) Raft tries to send a snapshot first.

Storage should provide an API to Raft to send a snapshot to a given node.