Two-column mode / OCR on manually specified regions

[shreevatsa]

Will likely need this

[shreevatsa]

The reason for starting with OCR instead of a blank editor was to avoid having to deal with splitting already added pages. But it may be useful to instead add one page at a time (could choose OCR option on a per-page basis). Would also be better than firing off hundreds of requests to the Google Vision API.

[shreevatsa]

Possible UI:

• User clicks a button
• Check that current chunk does not span multiple pages
• Bring up a div (like a modal), which shows:
  • the page region image corresponding to the current chunk
  • slider (like here) to split the page in two
• User sets the position of the slider(s), and clicks "done"
• Call OCR again with the two parts of the page respectively, and form chunks.
• Add all resulting chunks to the doc, replacing original chunk.

[shreevatsa]

Example of a modal div: https://chatgpt.com/c/fd4eae00-9c84-4e0d-bb82-06bc9873207a modal.zip

[shreevatsa]

After chunk split (either LR or UD) is decided, we need to:

1. Delete the original chunk from the editor,
2. Run OCR on the new regions, get lines out of the results, and chunks out of the lines,
3. Add back these lines and chunks (with updated xmin and xmax) to the doc.

Currently the schema already has lines' xmin and xmax, so in principle no schema changes are needed, although maintaining words is a bit annoying (#21).

We need some code refactoring to be able to do (2) easily.