Run route B over the OCR markdown of a PDF set

[peterbjohnson]

Every ticket writes its tests before its code, and its pull request reports a run over a target set or a corpus folder with the number of flagged fields. See docs/plan.md. Write every run's output to a file under out/ before reading it.

In the corpus sweep, route B did not run on any PDF set: the sweep hands the PDF itself to
pandoc, which reads no PDF, so the MECH60014 PDF sheets report route B failed: Unknown input format pdf and the UCL_MechEng folder no filter: pandoc reads no sheet of this set. For a PDF, route B reads the Mathpix markdown that route A reads: the filter is
written from that markdown's tree and pandoc runs it over that markdown. The ticket is done
when the UCL_MechEng sweep runs both routes on both sheets and reports a comparison.

---

Workbench ticket t48.

Live sweep over UCL_MechEng with both routes (from the implementer's run, run outputs kept out of the commit)

Sheet	Questions	Parts	Fields	Agreed	Adjudicated	Flagged	Not verbatim	Tokens
Worksheet_1.pdf	12	13	63	16	26	3	1	81,259
Worksheet_2.pdf	5	10	40	6	0	4	0	10,351