Native multilabel support on forecasting classification with CatBoost and XGB

[jonasblanc]

Is your feature request related to a current problem? Please describe. Following #2765 XGBClassifierModel and CatBoostClassifierModel are not leveraging native multi-label classification from their respective sub-model, they are relying on MultiOutputClassifier. Describe proposed solution https://xgboost.readthedocs.io/en/stable/tutorials/multioutput.html https://catboost.ai/docs/en/concepts/python-reference_catboostclassifier_fit#y

Both library seems to have some kind of support for native multi-label prediction. This would need to be further investigated but from what I gather they both require some label encoding which might be done in _format_samples.