Belief-Dependent Macro-Action Discovery in POMDPs using the Value of Information