My feeling is that Kedro should have an internal DAGs already, thus it is possible to automatically release dataset that is no longer needed. I actually thought this is the default behavior but it seems not true for the moment.