Better TQDM output

[mariosasko]

This PR does the following:

• if dataset_infos.json exists for a dataset, uses num_examples to print the total number of examples that needs to be generated (in builder.py)
• fixes tqdm + multiprocessing in Jupyter Notebook/Colab (the issue stems from this commit in the tqdm repo: tqdm/tqdm@f7722ed)
• adds the missing drop_last_batch and with_ranks params to DatasetDict.map
• correctly computes the number of iterations in map and the CSV/JSON loader when batched=True to fix tqdm progress bars
• removes the bool(logging.get_verbosity() == logging.NOTSET) (or simplifies bool(logging.get_verbosity() == logging.NOTSET) or not utils.is_progress_bar_enabled() to not utils.is_progress_bar_enabled()) condition and uses utils.is_progress_bar_enabled to check if tqdm output is enabled (this comment from @stas00 explains why the bool(logging.get_verbosity() == logging.NOTSET) check is problematic: [logging] unable to turn off tqdm logging transformers#14889 (comment))

Fix #2630

[lhoestq]

This is awesome thank you ! Looking forward to see this in action

It helps a lot to understand what's happening exactly, we have many progress bars now x)

[mariosasko]

@lhoestq I've created a notebook for you to see the difference: https://colab.research.google.com/drive/1by3EqnoKvC2p-yKW4lPDGOFOZHyGVyeQ?usp=sharing.

Feel free to suggest better descriptions for the progress bars.

If everything looks good, think we can merge.

[lhoestq]

That's awesome ! Love it :)