LANTERN: Scalable Distillation of Large Language Models for Job-Person Fit and Explanation