TransfQMix: Transformers for Leveraging the Graph Structure of Multi-Agent Reinforcement Learning Problems