feat: add support for masks in regrid filter

[b8raoult]

Description

Create a mask with

 anemoi-transform \
      make-regrid-file global-on-lam-mask \
     --global-grid era5.grib \
     --lam-grid carra.grib \
     --output era-over-carra-mask.npz

Then use it in a dataset recipe

input:
    pipe:    
       mars:
           class: ea
           param: [z, t, u, v]
           levelist: [1000, 500]
      regrid:
           mask: /path/to/era-over-carra-mask.npz     

What problem does this change solve?

This is a feature that was requested to extract the grid points of. global grid (e.g. ERA5 on N320) over the area covered by a LAM (e.g the CARRA domain)

What issue or task does this change relate to?

Additional notes

As a contributor to the Anemoi framework, please ensure that your changes include unit tests, updates to any affected dependencies and documentation, and have been tested in a parallel setting (i.e., with multiple GPUs). As a reviewer, you are also responsible for verifying these aspects and requesting changes if they are not adequately addressed. For guidelines about those please refer to https://anemoi.readthedocs.io/en/latest/

---

📚 Documentation preview 📚: https://anemoi-transform--108.org.readthedocs.build/en/108/

[yoel-zerah]

in tests/test_regrid.py, in test_make_regrid_mask the size of the mask produced by global-on-lam-mask is check against a reference provided in "anemoi-transform/filters/regrid/ea-over-rr-mask.npz".
Unfortunately, this reference mask has the wrong shape: the mask tested is generated by cropping ERA5 to CERRA (under 38k grid points), but the reference mask contains all of the original 542k ERA5 grid points.

• How can we change the provided mask with another file ?
• How do we generate a new reference file ? At the moment the only way I can generate a mask is with this feature being tested...

[yoel-zerah]

Loading a grid with src/anemoi/transform/filters/regrid.py:84: in as_griddata fails - when tests/test_regrid.py::test_regrid_ekd is run for instance:

def test_regrid_ekd(test_source):
    era5 = test_source("anemoi-transform/filters/regrid/2t-ea.grib")
>   regrid = filter_registry.create("regrid", in_grid="N320", out_grid=[0.25, 0.25])
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.venv/lib/python3.12/site-packages/anemoi/utils/registry.py:334: in create
    return factory(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^
src/anemoi/transform/filters/regrid.py:122: in __init__
    self.interpolator = make_interpolator(
src/anemoi/transform/filters/regrid.py:448: in make_interpolator
    return globals()[interpolator](**kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
src/anemoi/transform/filters/regrid.py:164: in __init__
    self.out_griddata = as_griddata(out_grid)
                        ^^^^^^^^^^^^^^^^^^^^^
src/anemoi/transform/filters/regrid.py:84: in as_griddata
    return lookup(grid)
           ^^^^^^^^^^^^
src/anemoi/transform/grids/named.py:114: in lookup
    npz = np.load(BytesIO(data))
          ^^^^^^^^^^^^^^^^^^^^^^
  
ValueError: This file contains pickled (object) data

The problem is that the anemoi.transform.grids.named.lookup function (or the equivalent deprecated anemoi.utils.grids.grids) called in anemoi.transform.filters.regrid fail at loading an npz file at e.g. url="https://sites.ecmwf.int/repository/anemoi/grids/grid-0p25.npz":

response = requests.get(url)
data = response.content
npz = np.load(BytesIO(data))

However, downloading manually the file and saving to path/to/grid-0p25.npz, and then loading with np.load("path/to/grid-0p25.npz") works.

Does anyone know how to fix this issue ?
**EDIT: This issue only occurs for some specific files, i.e. grid-0p25.npz, whereas it runs correctly for others grid-o96.npz **

[yoel-zerah]

I have finished reviewing all your comments, and I would like to request some help to finish closing this PR.
I am able to solve the merge conflicts by myself, but I don't have an answer for the currently unresolved conversations, could someone have a look ?