[model_card] Keep the model card readable even with many datasets

[tomaarsen]

Hello!

Pull Request overview

• Keep the model card readable even with many datasets
• Avoid 2 empty consecutive newlines in the model card

Details

This PR implements 2 new rules for the model cards:

1. If the ", "-joined string of dataset names exceeds 200 characters, then we just use trained on X datasets. with X as the number of datasets, instead of just trained on the aaa, bbb, ccc, and ddd datasets.
2. If we have more than 3 training or evaluation datasets, we add <details><summary>AllNLI</summary> and </details> around the dataset details.

cc @avsolatorio You don't have to do anything here, but you might be interesting in knowing that this will change soon.

• Tom Aarsen