Count keys in range

In a similar vein to #5789, it would be useful if it was possible to efficiently count the keys in a given range without having to load either the keys or values into the client.

I imagine this could be implemented particularly efficiently by partitioning the range into one subrange per storage server, sending a single request to each server, and then adding up the results.

Start by reading issue #5789 for the related design, then trace how range requests move between the client and storage servers. The issue’s proposed direction is one subrange request per storage server with results combined client-side; done means an efficient range-key count without loading keys or values into the client.

Written by the indexing model from the issue text.