Implement Celery for training models

[tk-king]

Currently, ml models are trained using background_tasks in fastapi.

This allows for scalablity and also takes the load from the ml-microservice.