Benchmark optimizations

[tomwhite]

The optimizations in #339 need to be benchmarked so we can see the effect they have.

The quadratic means workload that @TomNicholas benchmarked in https://xarray.dev/blog/cubed-xarray (notebook: https://gist.github.com/TomNicholas/c6a28f7c22c6981f75bce280d3e28283) is a good candidate since it has multiple inputs, and also a reduction step.

[tomwhite]

Work is ongoing in https://github.com/cubed-dev/cubed-benchmarks

[tomwhite]

Closing this as we are now collecting data at https://github.com/cubed-dev/cubed-benchmarks