Scaling-out comparison with Hail

[jeromekelleher]

We have a pretty good story on scaling up computation within a single server in our current "scaling properties". However, we don't have any concrete data on how we scale out. I think we do need one example of a computation that can be done on real data, and how it scales out.

We could do something like PCAs on 1000G data (in the cloud)? It doesn't have to be extensive, we just have to show that we can scale out, and that we're competitive with Hail.

We can refer to Liangde's Thesis (which is definitely citable) for a thorough comparison (which basically says sgkit is comparable with Hail, with a few caveats I think).