Transition `PyArray_PythonPyIntFromInt` to modern behavior

[seberg]

PyArray_PythonPyIntFromInt existed to not change behavior of existing code that used the "i" format in Python.
But along the way, Python fixed their things, the conversion now uses __index__ as it should and rejects all non-integers (not just floats explictly).

We missed that and while we used to be ahead of the curve we are now behind here. So what we should do is:

• Try the __index__ style conversion. (this is a tiny bug fix as well)
• If that fails, try current code with a warning.

It would be nice to audit if there are remaining "i" uses that matters, because if there aren't we can be clearer about when the warning was introduced (i.e. we won't use it for more functions).

Additionally, it may be nice to do a check whether all integers are used in places where NumPy may just want to directly transition to our integer conversion helpers. These differ in that NumPy sometimes rejects booleans.
(This doesn't matter much either way, the bool rejecting matters mostly in places that are slightly indexing related.)

[AbhijeetKushwaha1213]

Hi, I’d like to work on this issue as a standalone open-source contribution.

I’ve reviewed the current PyArray_PythonPyIntFromInt behavior and the transition described here. I’d like to investigate the relevant integer conversion paths, update the implementation to try the modern __index__-style conversion first, and preserve the existing behavior through an appropriate warning where compatibility requires it.

I’ll also audit the remaining "i" format usages and add or update tests to verify the new conversion behavior and compatibility cases.

Please let me know if I can take this issue.

[mattip]

@AbhijeetKushwaha1213 please do not post agentic ai comments here. Rephrase the work in your own words. We prefer to interact with humans. We do not assign issues, feel free to submit a PR after carefully reading our developer workflow and ai policy documentation.