Learning Visual Quadrupedal Loco-Manipulation from Demonstrations