IReCa: Intrinsic Reward-enhanced Context-aware Reinforcement Learning for Human-AI Coordination