Add proper tests with fill_value for test_interpolate_pd_compat-tests

[kmuehlbauer]

What is your issue?

In #8125 breaking tests in Upstream-CI was reported. Root cause is upstream #pandas-dev/pandas#54920 which is tackled in pandas-dev/pandas#54927.

To get the tests working, the fill_value-kwarg has been removed in #8139. The tests didn't cover the use of fill_value for scipy based interpolators (the used np.nan is the default-value) completely.

This issue is to keep track of adding proper tests for fill_value.

[kmuehlbauer]

@jhamman I'm trying to add proper tests for the above and asking myself, why we special case fill_value=None setting to np.nan for the left/right values:

xarray/xarray/core/missing.py

Lines 94 to 104 in c537238

	if fill_value is None:
	self._left = nan
	self._right = nan
	elif isinstance(fill_value, Sequence) and len(fill_value) == 2:
	self._left = fill_value[0]
	self._right = fill_value[1]
	elif is_scalar(fill_value):
	self._left = fill_value
	self._right = fill_value
	else:
	raise ValueError(f"{fill_value} is not a valid fill_value")

Is there a reason to deviate from the numpy.interp default values for left/right? I'm trying to get behind it, but can't think of any.

[kmuehlbauer]

Is there a reason to deviate from the numpy.interp default values for left/right? I'm trying to get behind it, but can't think of any.

Probably to align with the scipy interpolators which have a default fill_value=np.nan.