Text this: Adaptive Multi-Agent Reinforcement Learning with Graph Neural Networks for Dynamic Optimization in Sports Buildings