Risk-Sensitive RL: Near-Optimal Risk-Sample Tradeoff