Modeling rare events and nonmonotone nonignorable missingness of time-varying outcomes and predictors in binary time-series daily diary data: A Bayesian selection model