Combinatorial Bandits for Incentivizing Agents with Dynamic Preferences