Support chunking of returned datasets

[pont-us]

Currently, the CDS Store does not specify any chunking when opening downloaded NetCDF files. This means that the entire datasets are always loaded into memory (see open_dataset and open_mfdataset documentation). The store should provide some option to chunk datasets. Mirroring NetCDF chunking would be a good start, but see caveats in this discussion.

[forman]

For implementation refer to https://github.com/bcdev/nc2zarr/blob/main/nc2zarr/opener.py#L91, method _prefetch_chunk_sizes().