Extract-model

What's New:

* Support for SELFE datasets is now incorporated into `em.sub_grid()` `em.sub_bbox()` and
`em.filter()`.
* Support for numba and numpy implementations of `index_of_sorted()`.

What's New

* `extract_model` has a backend that will support reading in FVCOM model output which has previously not been possible when using `xarray` without dropping the vertical grid coordinates.
* `em.sub_bbox()` supports subsetting FVCOM model output.
* A new jupyter notebook demonstrating subsetting of FVCOM model output is now available in docs.
* `em.sub_grid()` supports subsetting FVCOM model output.
* `em.filter()` will not discard any unstructured coordinate information in the auxiliary coordinate
variables.

What's New
* `em.sel2d()` now uses `xoak` to find the nearest neighbor grid point on ND grids. Due to this change, `em.argsel2d()` doesn't exist anymore. Note that vertical functionality that was previously in `em.sel2d()` is now in `em.selZ()`.
* Provide more options in `em.filter()` for keeping different coordinates in a `Dataset`.
* Improvement to unit test setup.
* `em.preprocess()` will implicitly assign horizontal coordinates longitude and latitude for POM
datasets, even if the data do not specify `coordinates` attributes explicitly.

Hopefully improved reliability of tests running on CI. Also separated out `xESMF` into another installation file so that package will be installable from conda-forge for all operating systems, but then `xESMF` is installed subsequently for non-Windows users with the `conda-requirements.txt` file.

extract_model` can now be installed on Windows. If installing on Windows, horizontal interpolation will not work since `xESMF` will not work properly since `ESMPy` cannot be installed on Windows. Tests will run on Windows because those that use horizontal interpolation will not error out.

filter: can be used to easily filter a Dataset down to certain variables by their standard names, but also looks for the variables necessary to decode vertical depth coordinates.

sub_grid hadn't been written to account for both cases of dimensions and coordinates with the same name vs. not. Now it works for both scenarios.

Also cf-xarray is now available on conda-forge with the necessary changes so can integrate that.