Add support for HF transformers Donut

[WaterKnight1998]

This model is very difficult to use in production, it would be really appreciated if you add it into your library :)

[guillaumekln]

Thanks for the request.

We don't plan on integrating vision models at the moment.

I'm keeping this issue open since it could fit in the scope of the project (we just added a speech-to-text model), but don't expect the feature to come any time soon.