MAINT: more NumPy API shims

[tylerjereddy]

• to keep up with NumPy 2.0 related changes: API: Cleaning numpy/__init__.py and main namespace - Part 3 [NEP 52] numpy/numpy#24376

• this cuts us down to 36 failures, 128 errors vs. 48 and 129, respectively, on main

[skip circle]

[andyfaff]

I was going to add an array-api helper function that would do something like asfarray today. I think some of these casts are slightly incorrect. asfarray converts to a floating point type. If the existing type is already floating point type it doesn't do anything. The replacement here will convert float32 to float64, which we don't want. The helper function would find a few uses in optimize, it doesn't need to be public, unless we want it to.

[tylerjereddy]

Alright, sounds good, the testsuite isn't sensitive to the type width preservation, so we may want 1 or 2 checks for that if it is important then.

[tylerjereddy]

I revised this PR to back out the asfarray shims--I don't think it is sane for me to work on adjustments that the testsuite is not sensitive to.

The remaining changes here provide the improvements described in the revised original comment, and should directly match suggested substitutions at: https://numpy.org/devdocs/release/2.0.0-notes.html#numpy-2-0-python-api-removals

[andyfaff]

Do you still have the asfarray shims around?

I was going to add a helper function to act as an internal replacement, but I'd like to know how you dealt with removal of the function from __all__.

[andyfaff]

This looks ok to me. Going to merge in an attempt to get CI unstuck.