Learning a Unified Policy for Position and Force Control in Legged Loco-Manipulation