Fix PDF annotation matching across Markdown and line wraps

[tenglvjun]

Summary

• normalize parenthesized LaTeX relations and escaped Markdown percentages against PDF text
• resolve lexical and discretionary line-end hyphens independently for each occurrence while preserving exact source rectangles
• retain conservative ambiguity handling across forward matching, stored sort indexes, and folded context disambiguation

Cache compatibility

• bump the PDF text index profile from text-v8 to text-v9 so existing indexes rebuild under the updated normalization rules

Verification

• replayed the three reported NExT annotations against the real cached PDF index: pages 4, 5, and 6 with 11, 9, and 8 rectangles
• npm run check
• npm test (1168 passing)
• npm run build