fix(pdf): bridge unassigned decimal digits in native span fill

[kabishou11]

Problem

Native PDF span fill can drop decimal digits (ASCII or fullwidth Nd) when mixed CJK/digit fonts place those glyphs slightly off the layout span mid-line. Body text then loses the numeric run while nearby punctuation is still recovered, producing strings such as 至 ． 个 / 大约 个百分点. OCR mode is unaffected because it does not use this native assignment path.

This matches the CNKI-style field report in #5330.

Root cause

After primary tight/loose calculate_char_in_span assignment, _bridge_unassigned_punctuation recovered only Unicode category P* glyphs, and only when both visible neighbors already belonged to the same span and the glyph center was still inside that span.

Decimal digits (Nd) that fail the center-axis height-ratio check were never bridged, so chars_to_content never saw them. The same gap's punctuation could still be recovered.

Fix

Extend the existing neighbor-agree + center-inside-span bridge to single-codepoint decimal digits (unicodedata.category(ch) == "Nd") as well as P*. Safety gates are unchanged:

• both previous/next visible owners assigned and equal
• center inside the owner bbox via tight-then-loose
• no cross-line bridging (CONTROL_LINE_BREAK_CHARS still clears neighbor owners)

Letters (L*) are not bridged. calculate_char_in_span thresholds are not loosened. The helper is renamed to _bridge_unassigned_inline_glyphs / _is_bridgeable_unassigned_char.

Test evidence

Focused tests in tests/unittest/test_pdf_native_script_detection.py:

• fullwidth 至０．７３个 with CJK neighbors that pass geometry and intervening ０．７３ that fail calculate_char_in_span now restores 至０．７３个 (failed as 至 ． 个 before the fix)
• same geometry with ASCII 至0.73个 restores 至0.73个 (failed as 至 . 个 before the fix)
• unassigned digit between two different spans is not absorbed
• existing punctuation bridge tests remain green

.venv/bin/python -m pytest -q -o addopts='' \
  tests/unittest/test_pdf_native_script_detection.py \
  -k 'bridge or fullwidth or digit or span_fill_bridges or span_fill_does_not_bridge'

9 passed (2 failed before the production change).

Related to #5517
Related to #5330

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[kabishou11]

I have read the CLA Document and I hereby sign the CLA