Use LightOnOCR native model classes
Add torchvision dependency
Deploy LightOnOCR Space
initial commit
