rfftn/fftn shape mismatch with NumPy when s is specified?

[leofang]

xref: data-apis/array-api-compat#78 (comment)

>>> import numpy as np
>>> x = np.zeros((3, 2))
>>> np.fft.rfftn(x, s=(3, 2), axes=(1, 0), norm='ortho')
array([[0.+0.j, 0.+0.j, 0.+0.j],
       [0.+0.j, 0.+0.j, 0.+0.j]])
>>> np.fft.rfftn(x, s=(3, 2), axes=(1, 0), norm='ortho').shape
(2, 3)
>>> import cupy as cp
>>> x = cp.zeros((3, 2))
>>> cp.fft.rfftn(x, s=(3, 2), axes=(1, 0), norm='ortho').shape
(2, 2)
>>> cp.fft.rfftn(x, s=(3, 2), axes=(1, 0), norm='ortho')
array([[0.+0.j, 0.+0.j],
       [0.+0.j, 0.+0.j]])

and a similar issue exists for fftn.

Originally posted by @asmeurer in data-apis/array-api-compat#78 (comment)