Implementation of LearningManager and refactoring ExtractorManager into PredictionManager

[fmaschhur]

In order to efficiently parallelize model training and organization we need to implement a LearningManager class, which orchestrates model training and organizes incoming data into batches for online training of the models. Further, the ExtractorManager class needs to be refactored into a PredictionManager class. Currently, Learning is handled by Extractors in an unsuited way.

The implementation of the PredictionManager class could begin with copying the ModelRegistry of the instantiated ExtractorManager. Current state of development is available in the features/delta_train branch.