Added predictions for classification task. Fixed the check for normalizer presence.

[karannb]

• Updated make_ensemble_predictions to also predict for classification tasks, I am not sure if we need the aleatoric standard deviation in the output DataFrame.
• The check -

if "normalizer_dict" in checkpoint:

was incorrect, it would trigger even for classification tasks, because they have an empty normalizer dict for each task, but the attribute is present.

• Performed a sanity check on denormalization by using my code and a version without this logic + denorming manually, they both produced the same results (switched to the state at 5afeb6a before doing this experiment).

[karannb]

Continuation of #91.

[CompRhys]

thanks!