Gravar-mail: A Bayesian transition model for missing longitudinal binary outcomes and an application to a smoking cessation study