Draft file explorer

[Wauplin]

Basic idea would be to handle a directory path the same way we are handling a zip file. The abstraction let us navigate in the directory/zip file, check if subpath exists, is a file, is a dir, etc. Also allows to read the file as text or as bytes (lazy loading).

Goes together with the DDUF PR + this transformers' PR: huggingface/transformers#35173

Nothing is set in stone. Naming, specs, exposed methods, etc., everything can be changed/updated if needed. To be honest, I'm still very unsure this is a good idea. I find it elegant in theory but likely to be error-prone in practice + will require breaking changes in transformers. To be discussed...

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[Wauplin]

It was fun for the science but let's go for a simple solution 😄

(discussed offline but changing stuff in transformers for the sake of diffusers' DDUF is meeh 🤷‍♂️ maintenance will be a huge burden, been able to load from zip affects many internal methods for low impact, it might bring breaking changes and many untested variations => let's drop it and let diffusers do the rest. Worth case we unzip the transformers's components before loading them. We want a v0 to ship, not a perfect integration aux petits oignons)