Guard against a None video processor class when the backend is unavailable (#48557)
Without torchvision every VIDEO_PROCESSOR_MAPPING value is None while the
config type is still a key, so
`(video_processor_class or VIDEO_PROCESSOR_MAPPING[type(config)]).__module__`
raised AttributeError for models whose video processor cannot be inferred,
such as InternVL and InstructBLIP. That happens before the
is_torchvision_available() check further down, so the informative error added
in #44125 was never reached.
Mirror image_processing_auto.py: guard the None and compute
explicit_local_code only inside `if has_remote_code:`, the only place it is
used.