decouple mooring_array from cutout + bug + refactor + ...

[Mikejmnez]

Closes #378 #379 #384

Checklist

• The PR addresses an open issue.
• All tests are passing locally with pytest.
• The project passes linting with black and flake8.

Changes
This is essentially Part 2 of work I began back in March, when I started to work on mooring_array to allow compatibility with xoak. Here, a substantially more challenging and overwhelming effort which I began in July, decouples mooring_array from cutout, and addresses various open issues. In particular it closes: #378 , #379, #384 . See also #398.

To enumerate the features / fixes:

1. Allows mooring_array to be calculated without going through cutout first, whenever the extra argument serial=True is passed.
2. serial=False is the default behavior, and the new code implementation is separate from the previous way mooring_array works. This allows legacy code to remain intact.
3. Unpins scipy, and as a result OceanSpy will fail to build a tree when there is nan-data in the coordinates. This can happen with cutout + llc-data at high lats, resulting in an Error.
4. Because of 3, mooring_array + serial=True should be preferably used when face is a dimension.
5. mooring_array and also returns the pair (diffX, diffY), which is necessary for mooring_volume_transport. These variables are otherwise computed in mooring_volume_transport in a way that is incompatible with datasets with face as a dimension.
6. mooring_array + serial=True, computes the coordinates [XU,YU, XV, YV] from the coordinates of the mooring_array. Before, these were computed during cutout for the entire area encompassing the mooring_array, which was extremely inefficient. These 4 variables are required to be present in the dataset when computing mooring_volume_transport.
7. Significantly improves coverage.

In addition, this PR also refactors quite a bit subsample, although I had to create a bunch of functions and added them to llc_rearrange.

To see some of the behavior, I uploaded a testing notebook to GH, that is rather clean. The link is here: https://github.com/Mikejmnez/Notebooks/blob/main/ospy_mooring_latest.ipynb

Also, see this notebook for a comparison between the behavior of mooring_array via cutout (serial=False) and the new implementation (serial=True) with ECCO, one single snapshot.
https://github.com/Mikejmnez/Notebooks/blob/main/Mooring_volume_transport.ipynb

[Mikejmnez]

hmm I though adding shapely into the ci/environment.yaml would be enough...

[Mikejmnez]

hmm. I had to add shapely as a dependency in pyproject.toml.

[Mikejmnez]

this is taking way to long... I am going to restart the build of 3.11

[Mikejmnez]

Good, finally. FYI test_opening_and_saving within oceanspy/tests/test_open_oceandataset.py often times will get stuck, that is what was happening here, and it often happens in my local computer.

[MaceKuailv]

This is a lot of work and is really really cool!

[Mikejmnez]

Feel free to merge this. I introduced some functions in llc_rearrage to be used with mooring_array/stations that I think can and should be used in the arctic transformation, i.e. they'd greatly improve the behavior of the transformation. And so both llc_rearrage and its testing needs refactoring just like @MaceKuailv says. But I don't think I have the capacity for this now...