Improve error handling for invalid eval results in model cards

[hanouticelina]

When listing models with list_models(cardData=True), some models have incorrectly formatted eval results which causes errors during iteration:

[364]if self.model_name is None:
--> [365]   raise ValueError("Passing `eval_results` requires `model_name` to be set.")
ValueError: Passing `eval_results` requires `model_name` to be set.

This is flagged in #2894 and in this internal slack message.

This PR does two things:

• Adds type validation for eval_results to ensure they're EvalResult objects.
• when ignore_metadata_errors is true, allows users to ignore metadata errors (this is the case when parsing card data in ModelInfo).

[julien-c]

what's weird is that this is already validated on the server side IIRC

[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.

[hanouticelina]

what's weird is that this is already validated on the server side IIRC

@julien-c it seems it's not the case, because here for example : https://huggingface.co/nozomuteruyo14/Diff_LoRA/blob/main/README.md, eval_results is a string

EDIT: doesn't exist server-side, but users can add a eval_results field manually in the metadata

[Wauplin]

suggesting a change so that we definitely get rid of these messy errors 😬

[Wauplin]

All good! Thanks for iterating :)