Human-in-the-loop Learning for Dynamic Congestion Games