PDF widow/orphan controls

The new PDF format is presented much closer to the HTML, and this is all good.

There is one thing that I've noticed that seems to be a regression: [widow/orphan control](https://en.wikipedia.org/wiki/Widows_and_orphans) seems to be non-existent, while previous PDFs seemed to handle them reasonably. Widow examples: pp 49-50, and 50-51. Orphan example: pp 44-45.

Start by comparing the current PDF output with the earlier PDFs around pages 44-45, 49-50, and 50-51, then trace the repository's PDF generation path. Done means the affected sections no longer produce the reported widow and orphan page breaks while the PDF remains consistent with the HTML.

Written by the indexing model from the issue text.