Fix type error for Coarsen reduction methods

[andrew-s28]

Description

The methods for the documented reduction operations on Dataset.coarsen() and DataArray.coarsen() were dynamically generated via injected methods that relied on code in several different modules. The dynamic generation of these methods also meant that type errors were raised when trying to access reduction methods, e.g., the following code example.

Minimal working example

import numpy as np
import xarray as xr

rng = np.random.default_rng(42)
data = rng.random((10, 10))

ds = xr.Dataset(
    {
        "z": (("x", "y"), data)
    },
    coords={
        "x": np.arange(10),
        "y": np.arange(10)
    },
)

# raises Object of type `DatasetCoarsen` has no attribute `mean` ty[unresolved-attribute]
ds_coarsen = ds.coarsen(x=2).mean() 

This PR moves all the logic for Coarsen into xarray.computation.rolling and attempts to mirror the implementation of xarray.computation.rolling.Rolling. It should result in no functional changes as far as I can tell and should make future maintenance slightly easier (not having to track the injection across files).

Checklist

• Closes Type checking fails for DataArrayCoarsen methods #8136
• Tests added (note: I don't think any tests needed to be added here, but maybe in a larger refactoring, see below)
• User visible changes (including notable bug fixes) are documented in whats-new.rst

PS: I think the xarray.computation.rolling module might be due for a refactoring and don't seem to have been touched a lot in recent years. There's also #8059 related to this module, which I didn't address here to keep the scope limited. Happy to discuss a larger refactoring effort if there's any interest 😄

[dcherian]

That's a nice improvement! Can you fix the mypy errors please?

We ccould explore move these to https://github.com/pydata/xarray/blob/main/xarray/util/generate_aggregations.py but I think part of the issue is how some of these aggregations are written in terms of others.