A Cortical Microcircuit for Region-Specific Credit Assignment in Reinforcement Learning