Subgradient and Duality Methods for Optimal Stochastic Control with Applications to Real Options Valuation