Implement DataTree in auto_regression

[veni-vidi-vici-dormivi]

I implement DataTree as data format for our _scen_ens functions.
Towards #106

• Tests added
• Fully documented, including CHANGELOG.rst

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 78.37%. Comparing base (9ebcd27) to head (fb1eae2).
Report is 11 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #570      +/-   ##
==========================================
+ Coverage   77.93%   78.37%   +0.44%     
==========================================
  Files          49       49              
  Lines        2986     3047      +61     
==========================================
+ Hits         2327     2388      +61     
  Misses        659      659              

Flag	Coverage Δ
unittests	78.37% <100.00%> (+0.44%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mathause]

I will need another pass but looks good so far.

[mathause]

Thanks

[mathause]

Do we actually support arbitrary lags? No: #164 but also ok to leave.

[veni-vidi-vici-dormivi]

Actually, we do:

mesmer/tests/unit/test_auto_regression.py

Lines 522 to 544 in 1481de9

	@pytest.mark.parametrize("lags", [1, 2, [2]])
	def test_fit_auto_regression_xr_1D(lags):
	data = trend_data_1D()
	res = mesmer.stats.fit_auto_regression(data, "time", lags=lags)
	lags = lags if not np.ndim(lags) == 0 else np.arange(lags) + 1
	_check_dataset_form(
	res,
	"_fit_auto_regression_result",
	required_vars=["intercept", "coeffs", "variance"],
	)
	_check_dataarray_form(res.intercept, "intercept", ndim=0, shape=())
	_check_dataarray_form(
	res.coeffs, "coeffs", ndim=1, required_dims={"lags"}, shape=(len(lags),)
	)
	_check_dataarray_form(res.variance, "variance", ndim=0, shape=())
	expected = xr.DataArray(lags, coords={"lags": lags})
	xr.testing.assert_allclose(res.lags, expected)

The test with lags = [2] only fits the lag 2, without also fitting 1.