Policy Gradient in Continuous Time