On Solving a Stochastic Shortest-Path Markov Decision Process as Probabilistic Inference