Unpin array api strict

[JJFlorian]

Closes #10427

In array-api-strict>2.4.1, arithmetic operations no longer accept NumPy arrays. By inverting the assert_equal variables, we use numpy's __eq__ instead of the array-api-strict __eq__

[welcome]

Thank you for opening this pull request! It may take us a few days to respond here, so thank you for being patient.
If you have questions, some answers may be found in our contributing guidelines.

[JJFlorian]

I don't understand why the mypy tests fail after these changes, so can't really come up with a way to fix this (except for giving an ignore to mypy :) )

[Illviljan]

Array api strict is now typed since 2.4.0: data-apis/array-api-strict#135

This is the erroring code:

xarray/xarray/tests/test_dtypes.py

Lines 9 to 18 in 58594f8

	try:
	import array_api_strict
	except ImportError:
	class DummyArrayAPINamespace:
	bool = None # type: ignore[unused-ignore,var-annotated]
	int32 = None # type: ignore[unused-ignore,var-annotated]
	float64 = None # type: ignore[unused-ignore,var-annotated]
	array_api_strict = DummyArrayAPINamespace

Redefinitions with different types is in general not a good idea for type safety. mypy choose the first one and goes from there. Since array api strict wasn't typed <2.4.0, the type was Any which made the except case valid too, not anymore though.

We usually just ignore in these optional import cases, examples are here:

xarray/xarray/core/types.py

Lines 47 to 50 in 58594f8

	try:
	from dask.array import Array as DaskArray
	except ImportError:
	DaskArray = np.ndarray # type: ignore[misc, assignment, unused-ignore]

[Illviljan]

I think array api strict had the right idea here.
It's not a good idea to assume different types of arrays will work with each other.

Should we instead explicitly convert to numpy?
actual = np.asarray(xp_arr + 7)