Error Bounds in Nonlinear Control Design via Approximate Policy Iteration