Zooming into Comics: Region-Aware RL Improves Fine-Grained Comic Understanding in Vision-Language Models — arXiv2