Support mixed data formats

[aulemahal]

Change Summary

• Made Assets.format optional
• Added format_column_name to ESMDataSource's init
• Format information is added a column on ESMDataSource's internal df. xarray_open_kwargs are generated independently for each assets (instead of only once).

Related issue number

• Fixes Mixed formats in catalog #414

Checklist

• Unit tests for the changes exist
• Tests pass on CI
• Documentation reflects the changes where applicable

To make this PR simple, I decided to use an hardcoded name for the new column (_data_format_). Is this dangerous? It could be a column name stored in another variable, or the series could be independent from the main dataframe? The latter seems dangerous too.

I added a sample collection that has 4 assets of CMIP6. 2 assets are from Google Storage (zarr) and the 2 others are local files. This has the bonus to also test that to_dataset_dict also works with remote and local assets within the same dataset.

[andersy005]

To make this PR simple, I decided to use an hardcoded name for the new column (data_format). Is this dangerous?

I think this is good enough (It's similar to what we used to do before the re-write in #368).

I added a sample collection that has 4 assets of CMIP6. 2 assets are from Google Storage (zarr) and the 2 others are local files.

👍🏽 you forgot to commit the file though.

[aulemahal]

Woops! And on the day I am working from home with the other machine shut down... I'll push them tomorrow.
EDIT I realized the machine was indeed reachable and not shut down!

[andersy005]

Looks great 👍🏽