Linear convergence of a policy gradient method for some finite horizon continuous time control problems