Generalize tutorials for audio and vision

[stevhliu]

This PR updates the tutorials to be more generalizable to all modalities. After reading the tutorials, a user should be able to load any type of dataset, know how to index into and slice a dataset, and do the most basic/common type of preprocessing (tokenization, resampling, applying transforms) depending on their dataset.

Other changes include:

• Removed the sections about a dataset's metadata, features, and columns because we cover this in an earlier tutorial about inspecting the DatasetInfo through the dataset builder.
• Separate the sharing dataset tutorial into two sections: (1) uploading via the web interface and (2) using the huggingface_hub library.
• Renamed some tutorials in the TOC to be more clear and specific.
• Added more text to nudge users towards joining the community and asking questions on the forums.
• If it's okay with everyone, I'd also like to remove the section about loading and using metrics since we have the evaluate docs now.

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[mariosasko]

Looks good already.

Maybe instead of removing the metrics docs abruptly, we can deprecate them for now (with a warning?) and redirect users to the evaluate docs.

Some nits:

[lhoestq]

Awesome ! And +1 about deprecating metrics instead of removing them, and redirecting to evaluate

CI fails are unrelated to this PR, you can ignore them