EAGLE: Enhanced Visual Grounding Minimizes Hallucinations in Instructional Multimodal Models