ENH: update __module__ in numpy.random module

[mtsokol]

This is a follow-up PR after #27716. It updates np.random functions and methods' __module__ to numpy.random.

The only items left with incorrect __module__ are cdef classes, as Cython doesn't allows changing it.

List from here: pytorch/pytorch#136559 (comment)

[mattip]

For future reference, the cdef problem should be fixed by cython/cython#5479

[seberg]

LGTM. For Python files, I would prefer using the set_module helper, because I was in the situation before that I did not want this to happen.
(Specfically, some doctest related utils fail to find the source due to it.)

But, for cython code those things should already be failing, so...

[da-woods]

Re: module for cdef classes. A good fix looks really hard. The basic problem is that type defines its own property for __module__ which ignores whatever you set in the class dictionary for static types.

However. For heap types it works:

# distutils: extra_compile_args = -DCYTHON_USE_TYPE_SPECS=1

cdef class C:
    __module__ = "a"

print(C.__module__)  # prints "a"

Is that any use to Numpy?

(I'm also not sure that the issue that @mattip linked is a hugely great fix. It'd fix self.__module__ but not cls.__module__

[seberg]

Is that any use to Numpy?

Not sure I would enable it just for this sake, but I suspect heap-types are in our future anyway, so I don't mind doing it (unless someone has a clear downside).

FWIW, I don't think allowing to set __module__ and __name__/__qualname__ is very necessary/reasonable for non-heap types.
What I might allow (if it is easy) is exactly this:

cdef class myclass:
    __module__ = "something"

Which would set tp_name to f{__module__}.myclass rather than f{module.__name__}.myclass.