Gravar-mail: Sample size considerations for stratified cluster randomization design with binary outcomes and varying cluster size