Towards Efficient Risk-Sensitive Policy Gradient: An Iteration Complexity Analysis