It is currently not possible to train a deployed model on an entire dataset because dataiku forces the user to specify a test set.
See this thread detailing dataiku's current limitations when training a model: https://community.dataiku.com/t5/Using-Dataiku/when-training-a-model-with-a-visual-recipe-does-datai...
However, especially when a model has not reached its performance plateau (which one can see using a learning curve), it is considered best practice to train a model on all available data after finding the best combination of hyperparameters.
Adding this feature in the train visual recipe would be highly appreciated.