GraphEQA: Using 3D Semantic Scene Graphs for Real-time Embodied Question Answering