Gravar-mail: Accounting for interactions and complex inter-subject dependency in estimating treatment effect in cluster randomized trials with missing outcomes