arxiv InternLM-XComposer: A Vision-Language Large Model for Advanced Text-image Comprehension and Composition