fix(pdf): do not embed a CFF font of only .notdef

[andiwand]

🤖 Generated with Claude Code

Summary

The OTS check of build-test failed on main after #922 added ReichaCompositionslehreB2Teil4.pdf to the test data:

FAIL .../ReichaCompositionslehreB2Teil4.pdf/document.html: odr-f1
     ERROR: CFF : Failed to parse Top DICT Data

The font is HiddenHorzOCl, the hidden OCR text layer of the scan. Its CharStrings INDEX holds only .notdef. OTS rejects a CFF table with fewer than two glyphs (charstring_index->count < 2 in ots/src/cff.cc), and it reports this as a Top DICT error. cff::wrap_to_otf now throws for such a font, so font_is_usable renders the text with a substitute. The font draws no glyph, so no pixel changes.

Reference output

This PR also regenerates the references for the stack #920 to #924, as those PRs said. #911 and #916/#917 did not regenerate theirs either, so the new references include those changes too.

	files	cause
public	34 + 4 resources/	the viewport minimum-scale (#911), the pdf selection style (#916/#917), the frontend scripts (#912 to #918)
private	1112 + 4 resources/	the same, plus the two CCITT scans and the five corpus files (#920), the Intermediate cover (#921), and the Reicha scan (this PR)

If I apply the known CSS changes to the old references, only the files that the stack and this PR explain still differ. No file is added or removed.

• OpenDocument.test.output 055e7ec, OpenDocument.test-private.output 509b819, both on main.

Test

• CffFontTest.WrapRejectsAFontOfOnlyNotdef.
• The full suite passes (1848 tests). check_fonts.py on the Reicha output: 0 rejected by OTS. The changed files compile with -Werror.