Remove duplicate normalisation in FFT methods and enable relevant tests

[Wetitpig]

Some of the code originates from the branch yifeng/fft_c2r and yifeng/fft_r2c_with_abi, and also intel/intel-extension-for-pytorch.

However, the problem of redundant normalisations in the _fft_c2c_mkl_out function has been fixed. Some loops have also been simplified.

[Wetitpig]

Due to resource constraints I have been testing only the "usual" use cases (i.e.

• torch.fft.rfft, torch.fft.rfft2 on XPU
• torch.fft.irfft, torch.fft.irfft2 on CPU

and vice versa), and the results seem to be consistent with those from the original pytorch run on CPU.

[CuiYifeng]

@Wetitpig Thanks for your PR. Since I have completed most of the debugging (seems these cases are not fixed in this PR) based on my old branches, fft_c2r and fft_r2c will be submitted by another PRs.
Glad to see the fixing of normalization. Changing this PR to modify this part may be a better choice. Thank you.

[Wetitpig]

Now targeting only the repeated normalisation in *_out functions in FFT, while enabling tests for FFT and also some code refactoring