fix: handle None paragraph style in DOCX extraction

[mmaudet]

Summary

• Guard against para.style being None in _extract_docx(), which caused ingestion to crash on certain DOCX files

Root cause

para.style can be None for paragraphs in DOCX files created by non-Microsoft editors or converted from other formats. The existing code only guarded against para.style.name being None, not para.style itself.

Fix

# Before
style_name = (para.style.name or "").lower()

# After
style_name = ((para.style.name if para.style else "") or "").lower()

Fixes #6