Saddlepoint Approximations for Continuous-Time Markov Processes ∗