Fix unstack

[juliettelavoie]

Pull Request Checklist:

• pre-commit hooks are installed/active in my local clone ($ pre-commit install)
• This PR addresses an already opened issue (for bug fixes / features)
  • This PR fixes #xyz
• (If applicable) Documentation has been added / updated (for bug fixes / features)
• If a merge request has been made in parallel to this PR in xscen-notebooks, it is merged and the submodules have been updated.
• HISTORY.rst has been updated (with summary of main changes)
  • Link to issue (:issue:number) and pull request (:pull:number) has been added

What kind of change does this PR introduce?

• xs.utils.stack_drop_nans/ xs.utils.unstack_fill_nan format the to_file/coords string to add the domain and the initial shape.
• The initial shape is added to the lat and lon attrs to be able to retrive it when it is time to unstack.
• Redifining lon and lat to fix bug in xarray (reset_index not resetting levels of MultiIndex pydata/xarray#6946)

Does this PR introduce a breaking change?

I don't think so. If you pass a string without {domain} or {shape} as before, the formatting will be ignored.

Other information:

[aulemahal]

Ça me semble une bonne solution!
Petit questionnement sur la généricité (?) à conserver?

[aulemahal]

Je sais pas à quel point c'était utile, mais l'ancienne version de cette fonction était indépendante des noms de dimensions. On utilise mask.dims.
Est-ce que ça serait à garder?
Si oui, je propose:

Suggested change

	original_shape = f"{len(ds.lon)}x{len(ds.lat)}"
	original_shape = "x".join(map(str, mask.shape))

et il faudrait rendre le reste du code générique aux dims de mask...

Et si non, il faudrait juste le dire dans la fonction qu'on s'attend à lat et lon.

[juliettelavoie]

ah ben oui ça serait bien de garder cette flexibilité là. Je vais faire les changements.

[juliettelavoie]

The changes I have added will fail if one of the dimensions of ds_stack have an unrelated attrs also named original_shape, but that feels very unlikely...