Java DOCX renderer: align Latin document layout with Word (14_Thesis_Chapter)

[shps951023]

Summary

This issue tracks the Java DOCX renderer session that brought 14_Thesis_Chapter to near-pixel parity with Microsoft 365 Word.

Metric	Before	After
14_Thesis_Chapter overall	0.9856	0.9932
14_Thesis_Chapter text similarity	0.9979	1.0
14_Thesis_Chapter visual	-	0.983
Pages	2/2	2/2
Full issue/docx benchmark average	0.7225	0.7224 (28 cases)
Unit tests	20/20	78/79 (issues66 CJK smoke test now passes; 1 pre-existing Vietnamese issue91 extraction failure remains)

Session Flow

1. Diagnose remaining image differences between minipdf, Word reference, and LibreOffice renders of 14_Thesis_Chapter p1: headings not bold, bullets missing, body lines wrapped wider than Word, table offset, and wrong page-break position.
2. Heading bold — port Word's effective bold: paragraph-mark pPr/rPr and paragraph style rPr (CTParaRPr / CTRPr).
3. Bullet markers — resolve numPr → abstractNum → lvl via XWPFNumbering; map Symbol PUA bullet (\uF0B7) to •.
4. Heading color — apply paragraph-style rPr/w:color (Heading2 blue 4F81BD) when runs carry no explicit color.
5. Width metrics — mirror DocxToPdfConverter.EstimateCalibrTextWidth/EstimateWrapTextWidth:
   • Calibri-default documents: Times table with -8.8 units/letter kerning for serif runs, Calibri ×0.977 otherwise.
   • Other defaults: wrap with actual TTF advances (PDType0Font#getStringWidth), since Word lays out with the same TTF metrics.
   • Center/right alignment positioned with actual font advances.
6. Paragraph flow parity — contextualSpacing collapse between same-style paragraphs, keepNext widow control (reserve heading + up to 3 follow lines), top-of-page baseline = margin + font ascent, spacing-before suppressed at page top, spacing-after not collapsed across tables.
7. Table parity — auto-width table indent -108 twips (Word default), cell vertical margin 0, table spacing tuned to Word/LibreOffice measurements.
8. Bullet line height — Symbol-font labels advance 16.7pt (factor 1.21) matching Word and LibreOffice.

All changes are in minipdf-java/minipdf/src/main/java/io/github/minisoftware/minipdf/internal/docx/PoiDocxRenderer.java.

Validation

• mvn -B -ntp -f minipdf-java/pom.xml test → 78/79 passing.
• Run-Java-VisualBenchmark.ps1 -Suite issue -Format docx -SkipReference → 0.7224 average; 14_Thesis 0.9932, Issue79 0.9911/0.991, issues66 0.9879, Support_Letter 0.9761.