A Two-stage Reinforcement Learning-based Approach for Multi-entity Task Allocation