Florence: A New Foundation Model for Computer Vision