Enhancing Semi-Supervised Learning via Representative and Diverse Sample Selection