Feature/support multiple lams to the Cutout class

[paulina-t]

Summary

This PR adds functionality to the Cutout class to support multiple Limited Area Models (LAMs), along with a global dataset. This enhancement allows for hierarchical cutouts across multiple LAMs, handling of overlapping regions and different resolutions.

Motivation

The primary motivation for this change is to allow the anemoi-datasets package to perform cutouts for multiple LAM datasets (such as MEPS and AROME-Arctic) simultaneously, increasing flexibility in data handling.

Key Changes

Multi-LAM support in cutout: Enhanced the Cutout class to support multiple LAM datasets, handling them hierarchically where overlapping areas are masked sequentially.
Grid handling and resolution support: Updated the grids to compute and return grid point counts for each LAM and the global dataset, supporting varying resolutions.
Hierarchical masking: Introduced hierarchical masking of latitude and longitude in overlapping regions for multiple LAMs. The LAMs are provided in a list, with the first element having the highest priority (i.e., it is not masked). LAM no. 2 contains all the points not in LAM no. 1, and so forth. The last LAM in the list has the points that are not in any other LAM, and the global dataset is masked by all the LAMs.

Testing

Notebook Testing: Created a new notebook, grids_multilam.ipynb, based on grids.ipynb, to plot and visually verify that the Cutout class handles multiple LAM datasets correctly. This includes testing that the cutout works for:
- LAMs with different resolutions
- LAMs that do not overlap
- LAMs entirely contained within the domain of another LAM
- LAMs with a coarser resolution than the global dataset
- Compatibility with thinning and cropping

Note that the notebook relies on datasets generated from the files grids*.yaml. I added these files for reproducibility, but you might not want to include them in the repository. Let me know if there are other datasets you prefer to use.

Dependencies

No new dependencies: This update builds on existing functionality in anemoi-datasets without introducing new dependencies.
Compatibility: Fully compatible with single LAMs as well.

Notes for Reviewers

Potential areas of impact:
- Due to the significant updates in grid handling and masking logic, reviewers may want to verify compatibility with other classes or functions that rely on Cutout.
- The behavior of the method collect_supporting_arrays may need review. I modified it to store all LAM masks and the global mask, but please let me know if this differs from the desired output.
Documentation: Please let me know if additional documentation is needed to explain the multi-LAM functionality or if a style change is needed.

[FussyDuck]

All committers have signed the CLA.