Create new sections for audio and vision in guides

[stevhliu]

This PR creates separate sections in the guides for audio, vision, text, and general usage so it is easier for users to find loading, processing, or sharing guides specific to the dataset type they're working with. It'll also allow us to scale the docs to additional dataset types - like time series, tabular, etc. - while keeping our docs information architecture.

Some other changes include:

• Experimented with decorating text with some CSS to highlight guides specific to each modality. Hopefully, it'll be easier for users to find and realize that these different docs exist! Will experiment with this in a different PR.
• Added deprecation warning for Metrics and redirect to Evaluate.
• Updated set_format section to recommend using the new to_tf_dataset function if you need to convert to a TensorFlow dataset.
• Reorganized toctree to nest general usage, audio, vision, and text sections under the how-to guides.
• A quick review and edit to the Load and Process docs for clarity.

[HuggingFaceDocBuilderDev]

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

[lhoestq]

Nice thank you !

We can also add the "Load text data" page, currently it feels weird to not have it ;)

In particular users can do load_dataset("text", data_dir=...) or load_dataset("text", data_files=...) and they can use grep patterns to select several files.

The "text" loaders has a few parameters. The main parameter is sample_by. By default sample_by="line", so one example = one line from the text files, but you can change it to "paragraph" or "document"

---

A bit unrelated, but I feel like those pages can also be grouped together in a "Dataset repository" section:

• Share
• Create a dataset loading script
• Create a dataset card
• Structure your repository

This way we can decouple "General usage" (how to use datasets) from "Dataset repositories" (how to create a repository)

[stevhliu]

Ready for review!

The toctree is a bit longer now with the sections. I think if we keep the audio/vision/text/dataset repository sections collapsed by default, and keep the general usage expanded, it may look a little cleaner and not as overwhelming. Let me know what you think! 😄

[mariosasko]

Cool!

Just one nit.

[lhoestq]

Thank you ! I think it's ok to leave the new sections uncollapsed though, as you want