Zero-Shot Reinforcement Learning on Graphs for Autonomous Exploration Under Uncertainty