Bgen to zarr (v2)

[eric-czech]

#16

This is a second iteration on #21.

Notes:

• The conversion to zarr is done via rechunker instead of through custom code and an intermediate store.
• This expects Xarray integration in rechunker, so it won't really be ready until Add rechunking for Xarray datasets pangeo-data/rechunker#52 is done. For now I pointed the build at my fork.
• There are two functions added: bgen_to_zarr and rechunk_bgen. I'd expect bgen_to_zarr to be less useful than rechunk_bgen, but it's there for better consistency with sgkit-vcf.vcf_reader. The main advantage of working with datasets rather than paths is that it's easier to attach custom sample/variant metadata (as is usually the case w/ plink/bgen) and have it run through the same rechunking flow.

[ravwojdyla]

I suggest we review/merge this before https://github.com/pystatgen/sgkit/issues/256. @eric-czech is this good for review?

[eric-czech]

@eric-czech is this good for review?

Yep, there are still a couple things that will need to change after pangeo-data/rechunker#52 but otherwise it is ready to go. I haven't tested this at scale like the old version, but I'll leave any problems with that off for later.

I also want to add an import guard for rechunker and make it an optional dependency, but I'll follow the other examples we have for it when I get there.

[eric-czech]

Hey @tomwhite, can you review this when you get a chance? I'd like to merge it here so that I can then work on merging the whole repo (as @ravwojdyla suggested).

[tomwhite]

This looks good to me.