Consider request load on range shard splitting

Does range shard balancing consider the request load or just the data size when splitting ranges? What happens if a small range receives a disproportionate amount of read requests? Will the balancer try to migrate them to even out the load?

While hashed sharding may work better in some cases, there are situations where you still want to co-locate “close” data on the same shard, but at the same time, consider the request load when splitting the range (at least to some extent).