MAINT: redefine PyArray_MAX/PyArray_MIN because they moved in NumPy

[tupui]

Closes #18966

[tupui]

@rgommers @ev-br this seems to fix the issue 😃

[rgommers]

This seems to make CI happy, but redefining PyArrow_MAX without #ifndef guards may potentially run into more trouble later. Given that this macro nor the code that uses it have anything to do with NumPy arrays, I suggest renaming the macro to _MAX_OF_TWO_VALUES or some such name.

(fun with obscure C, in C++ we could just have std::max).

[rgommers]

The PyArray_MIN one is different:

  thetype = PyArray_ObjectType(image, NPY_FLOAT);
  thetype = PyArray_MIN(thetype, NPY_DOUBLE);

that still deals with integer type codes, but I guess we can clean that up since PyArray_ObjectType is superceded since forever: https://numpy.org/devdocs/reference/c-api/array.html#c.PyArray_ObjectType.

Okay, maybe let's get this in to unbreak CI, and deal with the rest in a follow-up.