I recommend using a stack recipe to get both train and test in the same dataset. In the stack recipe, you can add a new column specifying the origin: "train/test". Then you would have a single transformation pipeline, until the ML model where you would specify the train/test split using filters on the origin column defined in the beginning.