Refactor to cater to different types of users + caveat

[SubhadityaMukherjee]

There are two kinds of users

• Ones who would just use an openml dataset with their own pipeline.
• Students/other researchers who want a simple pipeline, say classification or regression with some model and don't want to modify the training itself.

This would be made easier to just have a basic training api and a more advanced one.

The main issue :

• Right now, it is not possible to upload training data without using a task and run_model_on_fold . This means anyone who uses their own pipeline will not be able to upload their data to openml. Perhaps this will be circumvented in the future, but now it is not really possible.