Skip download step, just create xarray datasets #10
Description
The goal is to open the full dataset. Storing the individual files locally is unnecessary and is actually the slowest part of the process. (Because we may have many thousands of small files, in terms of storage this is essentially random writes and random reads, which is slow.) It's much more efficient to write the entire dataset from xarray as netcdf or zarr.
Support for downloading the individual files can probably be dropped completely.