a quick question: is `predict_rolled()`the right way to create in-sample (on data the model has already seen) predictions? just double checking, because it looks like predict_rolled() mutates the model, which I wouldn’t expect, if it was just for “inference”. thanks!:)