Implement fit of ScienceEnsemble and use in DVC

[nsorros]

At the moment the models trained for science ensemble are trained independently. Since there is a model class ScienceEnsemble the training should move inside fit which will simplify dvc.yaml as well. Note that save will also need to be implemented.

There are two ways for implementing that

• through the use of WellcomeVotingClassifier which would be ideal
• through training the classifier independently inside fit and saving them

The WellcomeVotingClassifier might not be possible due to some internals of sklearn but we should investigate whether it makes sense to adjust the code to ensure it works. tfidf-svm is already an sklearn pipeline so any problems will probably arise from BertClassifier