Running multiple processes/threads for the same prompt

[varshith15]

From initial tests we noticed that CPU is the blocker and GPU is usually free. So in embedded comfy client, if we can use processpool instead of threadpool and run the same prompt twice, i think we can optimize the GPU even more and get higher FPS

[varshith15]

To test this hypothesis i add a 50ms delay to a basic video passthrough workflow in loadtensor and when i run the prompt once i get 18 FPS and when i run it twice i get 30 FPS. (this is with threadpool)

Now the issue with workflows like dreamshaper etc that the checkpoint loading functions are not threadsafe, so need to figure that out