Making the optimizer more flexible

[schalkdaniel]

This is important for a later step when adding new optimizer like momentum or Nesterov accelaration.
todo:

• Optimizer needs to have a calculatePrediction member function to calculate the prediction correctly w.r.t. the optimizer
• Response class should not set the prediction by itself. The optimizer should control that
• The logger should also get the optimizer in logStep to be able to correctly log the out of bag risk
• Is it possible to write the optimize function for parent class and just define the calculateUpdate and so on
• The prediction Function needs to be adjusted if the structure of the model depends no the optimizuer!