MAINT: special.perm: adjustments for backward compatibility/consistency

[mdhaber]

Reference issue

scipy#20909

What does this implement/fix?

Before scipy#20909, perm accepted lists and arrays as long as (k > N) or (N < 0) or (k < 0) worked; scipy#20909 immediately began to raise an error for non-scalar input. This changes it to be more flexible about the input unless it doesn't make sense (e.g. lists with more than one element and strings no longer work).

Also, scipy#20909 deprecated integral floats. This reverts that change for consistency with the deprecations made to comb. For instance, this still works without warning:

special.comb(np.asarray([5]), np.asarray(10).reshape(1, 1), exact=True)

so this still works without warning

special.perm(np.asarray([5]), np.asarray(10).reshape(1, 1), exact=True)

I don't think this is what we want, ultimately, but I figure that perm and comb should behave similarly so that a decision about what to do about both can be made together.

Marked as draft because I'd appreciate your thoughts rather than just merging. This is not necessarily correct; it's just trying to bring up some points that might have been overlooked.

[j-bowhay]

At a cursory look these changes look good. I am going to struggle to find the bandwidth to look in more detail in the next couple of days. I'll merge as is to see if other reviewers are able to review/are happy with this before branching, and if not, I can look in more detail when I find the time.