Exploring the role of reinforcement learning in vision–language models for cardiovascular disease decision support