Per subspace replication and storage engine

This project would allow users to specify a replication policy and storage engine for individual key ranges. There are variety of uses for this feature. One example is to keep a section of the database using the memory storage engine which contains keys that are changing rapidly. Another example would be to replicate a range of keys that is read very heavily to many more storage servers to help balance the read traffic.

Proposed design:
* keep a map in \xff\x02 that tracks subspace replication desires
* data distribution will attempt to keep shard boundaries and replicas that match the desired state, and once the system matches completely a key will be written to the database which can be watched

Start by reviewing the proposed subspace replication map in \xff\x02 and the data-distribution behavior described in the issue. Identify the existing code for shard boundaries, replica placement, storage engines, and database watches; done means users can configure per-range policies and the system reaches and signals the desired state.

Written by the indexing model from the issue text.