Planning and Learning with Stochastic Action Sets