Policy Optimization with Linear Temporal Logic Constraints