Benchmark and Improve Performance of Zarr streaming out of wk

[fm3]

The zarr streaming routes served by the datastore and tracingstore are pretty slow. Is there potential for optimization?

One thing that is certainly slow is the tiny request unit of 32³ voxels. We might eventually offer sharding-style routes.

But that does not seem to be the only slow part.

The libs streaming client is also a suspect (lots of bounding box instantiations?)