The Maskability Index: Predicting Task-Objective Alignment in Pretrained Language Models