Enable numbagg in calculation of quantiles

[maawoo]

• Closes Aggregating a dimension using the Quantiles method with skipna=True is very slow #7377
• User visible changes (including notable bug fixes) are documented in whats-new.rst

Just saw your message in the related issue @max-sixty. This is what I came up with earlier. I also did a quick test, comparing the calculation with and without using numbagg for a dummy 3D DataArray. I was only wondering if the default usage of numbagg (given that it's available and method='linear') should be noted somewhere in the docstrings and/or the docs in general.

[max-sixty]

This looks good!

We should add this fixture to some quantile tests so we're testing with and without numbagg:

xarray/xarray/tests/conftest.py

Line 17 in 4b5c87b

@pytest.fixture(params=["numbagg", "bottleneck"])

.

Here's an example of it being used:

xarray/xarray/tests/test_rolling.py

Line 325 in 4b5c87b

self, da, center, min_periods, name, compute_backend

— by adding it to the function, it'll run the test with numbagg turned on & off.

[max-sixty]

Looks like some of the tests are failing:

• It's attempting to calculate on arrays with units — we can exclude those when we decide whether we use numbagg.
• The error messages don't match — we can adjust the regex to include the error that numbagg returns

[max-sixty]

@maawoo have been out for a few days but left one comment which I think will fix the tests (though it may need refining slightly).

If you're up for it, could you add a whatsnew?

Then we can merge, thanks!

[dcherian]

Thanks @maawoo

[max-sixty]

Excellent work @maawoo ! Thank you!