From Values to Opinions: Predicting Human Behaviors and Stances Using Value-Injected Large Language Models