persist model after parallel sampling

Issue: changes to the model object are not persisted when parallel sampling is used. This prevents users from implementing their own trackers of model performance.

Therefore, after parallel sampling, the models should be collected again in order to persist any changes to their internal states.