Good-for-MDP State Reduction for Stochastic LTL Planning