MLDTA an Ensemble-Driven Multimodal Model with Dynamic Fusion for Drug–Target Affinity Prediction