Reinforcement learning of 2-joint virtual arm reaching in detailed cortex simulation