BUG: Make n a long int for np.random.multinomial

[thalassemia]

Fixes #25061.

[1fish2]

Should the Python Interface file mtrand.pyi also declare multinomial's n argument long?

It's currently typed _ArrayLikeInt_co which seems way off.

[seberg]

It is of course not safe to change this to long, this was a correct bug fix for windows and huge arrays.

[thalassemia]

The other changes to npy_intp I agree with and understand. Changing the type of n to npy_intp is, as far as I can tell, unnecessary and moot because it is subsequently assigned to a long later on in the function and not directly used again.

[seberg]

ops, sorry, you are right, somehow I thought it was used for a size. So we can do that, still wondering a bit about the other changes away from long (and presumably more in _generator.pyx) which changed if the cython version is important.

[seberg]

Thanks, this doesn't fix the underlying issue, but it seems fine and rejects very large values that don't fit long but do fit intp, which is potentially a bit safer on windows. Small enough fix even for the legacy random, I would say.