MANZANO: A Simple and Scalable Unified Multimodal Model with a Hybrid Vision Tokenizer - Explained Simply | ArXiv Explained