Feature request: skip non-image files

Something like:

```python
for d in docs[traversal_paths]:
if d.text is None:
if d.uri and d.tensor is None:
d.load_uri_to_image_tensor()
self._reshape(docs[traversal_paths])
self._normalize(docs[traversal_paths])
```

I'd make a PR right now, but my brain is fuzzy from working on other stuff

This ties into my PDF search engine - I've got text chunks and image chunks on the same level. I need to skip preprocessing on all the text chunks.