Transformer-Based Scalable Multi-Agent Reinforcement Learning for Networked Systems with Long-Range Interactions