Additional afdist implementation benchmarks

[jeromekelleher]

We have additional implementations of afdist to benchmark

• Java: Java afdist #168
• Tensorstore C++/Python

As having all the different implementations on one figure would be quite messy, the simplest thing may be to add an additional supplementary figure which just compares the Zarr implementations. Then, we can do that on a different machine and hopefully sidestep the annoying problem in #161

@benjeffery, what do you think? I'm imagining we have a supplemental figure which looks like Fig 3 (whole matrix compute) but where we only run the Zarr implementations. That way it shouldn't take to long to do, and it'll let us showcase the strength of the Zarr ecosysem.