Feature/kwargs for zarr open dataset

[EddyCMWF]

1. open_dataset -> open_zarr for succinct kwarg passing
2. Option for parsing kwargs into open_zarr command
3. decrypt kwargs recursivley

[malmans2]

Approving, but with a couple of comments:

1. I was under the impression that xarray's long-term plan was to deprecate open_zarr. I cannot see a warning in the documentation though, so I may be wrong. It's no big deal if you are sure that you will always open all ARCO data with zarr-python.

2. I would have handled the decryption differently, by explicitly decoding only the credential arguments. The blind recursive decryption seems a bit error prone to me. I am also not a big fan of the ignore_errors option in decrypt. I only added it to comply with the requirements, but I would avoid using it when possible.

[EddyCMWF]

Thanks @malmans2 ,
For 1, I will change back, I was not 100% sure on that, and your suspicion is enough for me to revert. I can then make engine: "zarr" a default kwarg which can be over-ridden if need be in the future... not sure it ever will be

For 2, understood but happy to go with it as it is as the exact credential config is still being discussed and this gives us a nice flexibility. I will open an issue to come back in the future.

[malmans2]

Sounds good.
BTW, I think you might also want to add this default, which will default to Dask arrays.

open_dataset_kwargs.setdefault("chunks", {})

[EddyCMWF]

Sounds good. BTW, I think you might also want to add this default, which will default to Dask arrays.

open_dataset_kwargs.setdefault("chunks", {})

I'll investigate with configs then make default if needed