sym_rfftn for ifftn
This is an ommission from the changes I made in my scipy PR:
ifftn for real input can be optimised in exactly the same way as with
fftn. You just need to take the complex conjugate of the 1D r2c transform before continuing to the remainging axes.
I suspect it's not a very common use case but it's not difficult to add. Alternatively, this could go into pocketfft itself as a
forwards argument to