Towards Scalable Adaptive Learning with Graph Neural Networks and Reinforcement Learning