Improve handling of hot read shards

If a range of keys is being heavily read, the only mitigate that currently exists is that clients will start to load balance traffic to other shards when possible. This strategy does not help once a single range is flooded with enough read traffic to saturate the servers that hold that range.

The goal of the project is to make data distribution aware of hot read shards, and either split the range or increase the replication factor for the shards. This will allow more servers to participate in serving the reads.

The issue describes making data distribution aware of hot read shards. Start by understanding how the project currently distributes and replicates ranges, then determine whether splitting ranges or increasing replication is appropriate; done means hot read traffic can be served by more servers.

Written by the indexing model from the issue text.