Fix pipeline deadlocks, remove torch.compile, implement 3-queue parallel pipeline, optimize for 16fps

Reorganize setup files and update documentation

Initial commit: speech_to_video project with models via LFS