Reduced Policy Optimization for Continuous Control with Hard Constraints