Performance benchmarks

See that the GPU/CPU [pipelines](https://github.com/dkazanc/dls_pipelines/) run for various typical data sizes. 180 and 360 LFOV scans included.

I put benchmarks before merging everything into `gpuloop` as I think we discover few issues/bugs on the way, so it makes sense to resolve them. However, I also see that after the merge we can have different issues as well! So feel free to decide what you think the best order is: `test -> merge -> benchmark` or `benchmark -> merge -> test`. Arguably the former is more risky but potentially less work?