Rework train_if functionality in online learning

[benHeid]

The train_if argument is evaluated when the model is fitted, which happens before the transformation. However, refit should be applied after a transformation. Moreover, instead of using a function for train_if we should use a condition object, which gets the output of steps as input and no complicated workarounds has to be done by the user like storing the result of a FunctionModule in global variable which is then evaluated by the train_if function. See #116