Learning to Compress Graphs via Dual Agents for Consistent Topological Robustness Evaluation