Writing a sd-turbo training script that aims to provide fast single step high resolution images.

To do this we continue the work of Adversarial Diffusion Training for distillation by bringing in the STOTA GAN losses from the Giga-GAN paper.

Training Scripts: - [x] GAN losses from GigaGAN - [x] Teacher Loss - [x] Multi Aspect Ratio Training