Bandits atop Reinforcement Learning: Tackling Online Inventory Models with Cyclic Demands