docs: document schema for datasets in tasks

[MartinBernstorff]

When I added #247, it was non-obvious for me which columns a dataset should contain.

I propose documenting this at the AbsTask level, e.g. for classification something like:

class AbsTaskClassification(AbsTask):
    """
    Abstract class for kNN classification tasks
    The similarity is computed between pairs and the results are ranked. 
    
    Dataset must be a huggingface dataset split into train/test, and contain the following columns:
        text: str
        label: int
    """

[imenelydiaker]

I completely agree with you! We ran into the same issue when creating the benchmark for French. We should add a docstring like this one for each task type maybe 🤔

[MartinBernstorff]

I'd gladly take on part of this btw 👍

[KennethEnevoldsen]

Perfect @MartinBernstorff will assign it to you. Feel free to add me as the reviewer