Argus: Leveraging Multiview Images for Improved 3-D Scene Understanding With Large Language Models