HYPERmotion: Learning Hybrid Behavior Planning for Autonomous Loco-manipulation