[ ] Inspecting the bounding box code in the Table structure predictor, and making sure that the correct bounding boxes are being extracted/used for intersecting with tokens
[ ] Shortening the oddly long token boxes that PDFPlumber puts out, so that they're not as likely to interfere when doing bounding box things. This caused issues in #14 as well.
This involves two potential approaches: