`NamedArray` typing audit

[ilan-gold]

What is your issue?

See: #9671 (comment)

At the moment, NamedArray's backing data can fulfill _arrayfunction although in practice, PandasIndexingAdapter and PandasExtensionArray don't actually fulfill this contract (although the later is much closer than the former). The following two lines are particularly problematic for non-numeric data:

xarray/xarray/namedarray/_typing.py

Lines 188 to 192 in d57f05c

	@property
	def imag(self) -> _arrayfunction[_ShapeType_co, Any]: ...
	@property
	def real(self) -> _arrayfunction[_ShapeType_co, Any]: ...

FWIW numpy arrays do always fulfill this even when it doesn't make sense (even object dtypes have a imag and real). So I'm not sure how to move forward.

Of course, another issue is the whole _DType_co being bound to numpy. __getitem__ can return pandas dtypes but __array__ won't so maybe we need a base dtype and then a numpy one as well in the generic for the class?

[Illviljan]

For imag and real: #9368 (Been some iterations on this topic, #8295, #8365)

For the dtype issues I've made an attempt in #8933, but as usual it's not that easy to just switch to a simpler dtype class.

[headtr1ck]

For the dtype issue I guess we will run into the problem of not having higher kinded types.

Not sure how one could solve that without some ugly descriptors.