Optimal control with state constraints: a semiconcavity result