Gravar-mail: Super-Learning of an Optimal Dynamic Treatment Rule