Gravar-mail: A pattern-mixture model approach for handling missing continuous outcome data in longitudinal cluster randomized trials