Toward Reinforcement Learning-based Rectilinear Macro Placement Under Human Constraints