Proper backend support

[andreped]

This PR is the result of the discussion from #11.

The aim was to be able to use different backends, without the need to install all backends. This is useful as torch/tf often collides, which may hinder people to use staintools out-of-the-box.

What I did was to make subpackages within the main package, totally stratified into three packages: base, tf_backend, torch_backend. The numpy backend is part of the base, as both tf and torch depends on numpy. I also added more CIs to properly verify that it works, but more can be added if necessary. But it should be fine for now.

Last that remains is to simplify imports, as mentioned #11 (comment).

Milestones:

• Stratified package into three subpackages for the code base and the two backens.
• Added specific CIs to verify that the tool works for tf/torch, without the need for both.
• Simplify imports.
• Final refactoring when everything works as intended.

[carloalbertobarbano]

Thanks for the PR, I will merge it into a new branch and work on fixing the imports

[carloalbertobarbano]

Seems ok