Hierarchical DLO Routing with Reinforcement Learning and In-Context Vision-language Models