Fix async events for OpenAI API extension + Other small fixes

[Artificiangel]

Checklist:

• I have read the Contributing guidelines.

Changes

• OpenAI API extension now allows different tasks to run in parallel without locking up the webserver.

  • Run embedding functions while generating text (Originally only worked with streaming).
  • Send a request to StableDiffusion-WebUI while loading a text generation model.
  • Limit heavy tasks (split by category) to run in queue.
  • Tested all modified endpoints.

• When throwing an error: Fixed model/load and lora/load endpoints returning errors with a 200 status code.
  Being unexpected, that might not be caught by users on client side.

• For consistency added return "OK" to model/unload endpoint, matching lora/unload.

• Removed awaiting of sync OAIimages.generations function in the images/generations endpoint that caused an error.

• Use custom model/lora folder in downloader. (reverted to merge)

• Fixed transcription endpoint.

• Added Fastapi docs for: Transcriptions, image generation