Respecting Markov Equivalence in Computing Posterior Probabilities of Causal Graphical Features