Duality and upper bounds in optimal stochastic control governed by partial differential equations