Reduce memory usage of ctree

[JensWahl]

When explaining predictions using explain(..., approach = "ctree") we create one ctree model for each row in the binary matrix S. The training data is saved in each ctree model meaning that it's is saved nrow(S) times or nrow(S) / n_batches when n_batches > 1.

There should be a way to reduce the size of the ctree object by removing the training data while still being able to sample from the model.