Does not pass the ocr-fileformat validator yet (<tr> elements are
unexpected)

We can re-use it when parsing tables (each cell is a block)

Will move most of the code to a separate parsing file later

The hocr-fold-chars fix requires a version bump