Add information how to see the features of the dataset

[adam-narozniak]

Issue

The feature information (the name of the image label - either "img" or "image" that differs on HF) is needed to make decisions on the transformation of the dataset. Yet, it's not emphasized enough in the docs.

Proposal

Improve the documentation: quickstart + all the how-to-{framework} sections to include this information.

This PR also slightly restructures the TensorFlow example by pulling the common part to the beginning and not repeating it (there not repeating the newly added section neither)

Explanation

It is needed:

• to apply transforms in PyTorch,
• divided the dataset into X and y in NumPy,
• transform into TensorFlow Dataset in TensorFlow

[jafermarq]

I left some comments!

Although not part of this PR, i think the Use with PyTorch page would benefit from some sections/subsections just like you do for TF and NumPy