Conformal Policy Learning for Sensorimotor Control Under Distribution Shifts