Fix TypeError with empty axes in FFT functions

[antonwolfy]

This PR fixes "copyto() argument 1 must be a numpy.ndarray" error, like:

deploy/_test_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_place/lib/python3.13/site-packages/mkl_fft/_fft_utils.py", line 250, in _iter_complementary
    np.copyto(result[tuple(sl)], func(x[tuple(sl)], **kwargs))
    ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: copyto() argument 1 must be a numpy.ndarray, not numpy.complex128

when calling FFT functions with axes=().

Thus the PR adds a fast path to handle empty axes as identity transform, matching NumPy behavior.

The issue was introduced in scope of #283. The tests coverage is improved.

[jharlow-intel]

linux validation failure is due to bdba server outage, nothing to do with this PR.

LGTM, thanks for adding the new test cases, too

[ndgrigorian]

I spent a bit of time trying to figure out the cause of this change, eventually it made sense, but I realized that it would be easier if these functions in this file were better documented...

It's not in the scope of this PR to fix all of them, but would you mind adding one to this function? @antonwolfy in particular mentioning now that a is expected to be numpy array

[ndgrigorian]

other than previous comment LGTM