Visual Salamandra: Pushing the Boundaries of Multimodal Understanding