Learning Monocular Depth from Events via Egomotion Compensation