A Unified Framework for Variable Selection in Model-Based Clustering with Missing Not at Random