fix: emit glyph marker for glyph-index names instead of leaking them as text

[wittjeff]

Addresses the glyph-name half of #238.

Problem

Subset generators (FontForge, fontTools, mPDF, ...) name glyphs by bare index: /gid00043, /g43, /glyph43. Such a name identifies the glyph inside the embedded font program but carries no reading text. When /Encoding/Differences assigns these names and no /ToUnicode CMap covers the code, init_differences()'s last-resort fallback kept the raw name as reading text:

/Differences [ 65 /gid00043 /gid00049 /gid00041 ]  →  "gid00043gid00049gid00041"

Because that is plausible-looking text rather than a GLYPH<...> marker, downstream consumers cannot detect the corruption — docling's rate_text_quality() gate never fires and no OCR fallback triggers. This is the same fabricated-text class as #299, reached through the glyph-name path instead of the encoding path.

(Note: the with-ToUnicode variant of #238 is already fixed on main — init_differences() consults the CMap first per PDF 32000-1 §9.10.2. Released 5.11.0 still emits /gidNNNNN even when a valid ToUnicode exists. And the file attached to #238 does not reproduce for me on 5.11.0 or main — details in the issue thread.)

Fix

The last-resort fallback (resolve_unknown_name) now detects pure glyph-index names — (gid|glyph|g|cid|index)\d+, case-insensitive, full-match — and emits a GLYPH<name:gid00043> marker instead of the name. Under the default keep_glyphs=false the marker is stripped to a space, so no fabricated text reaches consumers; with keep_glyphs=true the glyph identity is preserved for diagnostics.

Deliberately unchanged:

• a valid /ToUnicode CMap stays authoritative ahead of the guard (method 1 of §9.10.2);
• meaningful unknown names (custom ligatures like /Th, /ft) keep the existing name-as-text fallback;
• glyph-name based methods (AGL tables, uniXXXX, ligature join/decompose, /C<digits>) all run before the guard.

Recovering the actual text for these fonts would require reverse-looking-up the embedded font program's cmap table (gid → Unicode) — worth discussing as a follow-up; this PR makes the corruption detectable so OCR fallback can kick in meanwhile.

Tests

tests/test_gid_glyph_names.py builds the PDFs in memory (same pattern as #299's tests): gid-named Differences with and without ToUnicode, name variants (g43, glyph49, index41), and a regression guard for ligature-style names. Full suite passes (88 tests).

[github-actions]

✅ DCO Check Passed

Thanks @wittjeff, all your commits are properly signed off. 🎉

[mergify]

Merge Protections

🟢 Merge protection satisfied — ready to merge.

Show 1 satisfied protection

🟢 Enforce conventional commit

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?: