Use Botorch MultiTaskGP for transfer learning

[Hrovatin]

This can stay a draft until it is validated that the performance improves with this model and that we even want to use it.

Things to resolve/issues:

• Posterior can not be computed for tasks that are not present in the measured data, meaning that recommendations can not be made for active task that has no existing data. Since in this case the values of the index kernel are anyway random, would it make generally sense to just use random recommender (or sth else that is not task-GP) if no target task points were measured, even if the the training task data is present? -> EDIT: This already works but not in with StratifiedStandardise.
• StratifiedScaler does not enable multi-output models, see [FEATURE REQUEST]: StratifiedStandardize for multi-output models pytorch/botorch#2739
• Acquisition functions currently do not support multiple active task values as this would create multiple posteriors.