A Biologically Plausible Architecture of the Striatum to Solve Context-Dependent Reinforcement Learning Tasks