Reinsert broadcasted mask

[FabianHofmann]

Sorry for the back and forth.

This reverts the FutureWarning from #579 and re-enables np.where (from #555) for applying masks, which indeed is significantly faster than xr.where.

The root issue was that DataArray masks weren't broadcast to match data.labels, unlike masks passed as pd.Series, pd.DataFrame, or np.ndarray (which go through as_dataarray broadcasting). This is now handled by a new broadcast_mask helper that calls broadcast_like to align/transpose the mask, then detects and warns about missing coordinates, falls back to NaN fill with False.

Dimension validation is now again a hard assert (mask.dims ⊆ data.dims) instead of a FutureWarning.

Checklist

• Code changes are sufficiently documented; i.e. new functions contain docstrings and further explanations may be given in doc.
• Unit tests for new features were added (if applicable).
• A note for the release notes doc/release_notes.rst of the upcoming release is included.
• I consent to the release of this PR's code under the MIT license.

[FabianHofmann]

@FBumann could you also have a look. I can't select you as a reviewer here (would need to add you as a maintainer, you want that?)

[FBumann]

@FBumann could you also have a look. I can't select you as a reviewer here (would need to add you as a maintainer, you want that?)

I'll review it this evening.
Thanks for your trust :). Maybe we can setup a call at some point and get to know each other, and then discuss this (briefly)?
I wanted to do that for some time with @lkstrp anyway. I'll write you an email.

[FBumann]

@FabianHofmann See #581 for small modifications. Even without: LGTM

[lkstrp]

Thanks for your trust :). Maybe we can setup a call at some point and get to know each other, and then discuss this (briefly)?

@FBumann We are happy to have a call to meet and align! Just send me an email again with some dates this or next week or whenever it fits you. @FabianHofmann will join as well