add FIT and other various of updates

• FIT
• FITDenoiser
• FITAR
• update on data pipeline to support data mixing
• ...

diffusion models and more tasks

• adding diffusion generative models
• support transforms as config
• support more tasks (image/video generation, panoptic segmentation)

initial pix2seq v2 release

Support multiple vision tasks in a single neural network (with different prompts), namely, object detection, instance segmentation, keypoint / human pose detection, image captioning.

some minor changes to original pix2seq release for pix2seq v1 paper

this is a snapshot before we release the code for pix2seq v2 (multi-task version)

initial release

Initial release of the pix2seq codebase for object detection task