Fixes NumPy interface bug which did not allow for list/ndarray axes or s parameters

[cako]

Description

NumPy multidimensional FFTs accept tuple, list and np.ndarray for the axes and s parameters. Upon substituting a code which ran on standard numpy for one running on intel-numpy, I came across a bug where axes and s struggled to handle anything apart from tuple.

This came down to a couple of bugs in mkl_fft which are fixed in this PR. In addition, I added a test which should catch these bugs in the future.

Edit: Commit 75e9ffd is unrelated to the bug. It's just a code improvement.

Reproducible examples

To simplify checking, here is a list of behaviors which work/do not work currently. This PR fixes them.

import numpy as np

np.fft.fftn(np.ones((2, 3, 4)), axes=(-1, 0))  # Works
np.fft.fftn(np.ones((2, 3, 4)), axes=[-1, 0])  # Does not work
np.fft.fftn(np.ones((2, 3, 4)), axes=np.array([-1, 0]))  # Does not work
np.fft.fftn(np.ones((2, 3, 4)), axes=(-1, 0), norm="ortho")  # Works
np.fft.fftn(np.ones((2, 3, 4)), axes=[-1, 0], norm="ortho")  # Does not work
np.fft.fftn(np.ones((2, 3, 4)), axes=np.array([-1, 0]), norm="ortho")  # Does not work

np.fft.fftn(np.ones((2, 3, 4)), s=(3, 4, 5))  # Works
np.fft.fftn(np.ones((2, 3, 4)), s=[3, 4, 5])  # Works
np.fft.fftn(np.ones((2, 3, 4)), s=np.array([3, 4, 5]))  # Works

np.fft.fftn(np.ones((2, 3, 4)), s=(4, 5), axes=(-1, 0))  # Works
np.fft.fftn(np.ones((2, 3, 4)), s=[4, 5], axes=[-1, 0])  # Works
np.fft.fftn(np.ones((2, 3, 4)), s=np.array([4, 5]), axes=np.array([-1, 0]))  # Works
np.fft.fftn(np.ones((2, 3, 4)), s=np.array([4, 5]), axes=np.array([-1, 0]), norm="ortho")  # Does not work

[oleksandr-pavlyk]

Thanks you @cako