DSCC-HS: A Dynamic Self-Reinforcing Framework for Hallucination Suppression in Large Language Models