add Nafnet model to ldct-benchmark

[kardSIM]

Summary

Hi, I like this benchmark and found it very useful. I wanted to contribute by adding NAFNet as a new LDCT denoising method.

Changes proposed in this pull request

• Add NAFNet as a new LDCT denoising method.
  Reference NAFet repository: https://github.com/megvii-research/NAFNet

Evaluation

I was able to obtain the Chest and Abdomen portions of the LDCT dataset, but I could not access the Neuro data because it requires restricted access. Therefore, I trained and evaluated NAFNet using only Chest and Abdomen data.

The obtained results are:

Method	PSNR (Chest)	PSNR (Abdomen)	PSNR (Neuro)	SSIM (Chest)	SSIM (Abdomen)	SSIM (Neuro)
LD	18.066	29.117	nan	0.312	0.856	nan
offline-run-20260812_235210-1ktgpkg2	28.092	33.939	nan	0.586	0.913	nan

I did not update the benchmark documentation, since I don't have access to the Neuro data and updating it would result in an unfair comparison. I would be happy to discuss the appropriate way to report these results.