Support multi labels textcat pipe for transformers

[tamuhey]

closes #9

• Add TrfForMultiLabelSequenceClassification for multiple text classification.
  • pipe name: transformers_multilabel_sequence_classifier
• Add docs for fine-tuning multi textcat pipe
  • https://github.com/PKSHATechnology-Research/camphr/blob/feature%2Fmulti-textcat/docs/source/notes/finetune_transformers.rst#multilabel-text-classification

[ghost]

How about implementing _labels method, mainly for multi-label classification?
The method name is up to you. _labels may be ambiguous.

def _labels(doc: Doc) -> List[Tuple[str, Any]]:
    if not doc.cats:
        return []
    return list(doc.cats.items())

[tamuhey]

It's not a very good way.
Since doc.cats contains all labels, it is same as using doc.cats directly.
You need to set a threshold to return high scoring labels, but I think this should be left to user.

[ghost]

Since doc.cats contains all labels, it is same as using doc.cats directly.

I agree with you! I'm sorry!

[ghost]

Is it possible to validate ".jsonl" for textcat (not multitextcat)?
To avoid the below case.

When trying multitextcat and forgetting to switch the argument from model.textcat_label to model.multitextcat_label, the training phase works as textcat(not multitextcat) even if the sum of labels is not 1.

[tamuhey]

Added data validator