fix(extractor): honor PDF text horizontal scaling

Merged
AB
2mo ago
+805-52

Summary

Honor PDF text horizontal scaling (Tz) when calculating text bounds and advances, including reflected text and nested Form XObjects. Preserve the scale through graphics-state changes and retain painted bounds when font size or horizontal scaling reflects glyphs inside ActualText, preventing scaled text from being assigned to unrelated regions.

Corrected coordinates can expose existing whitespace and layout inference edge cases in tightly spaced text; those heuristics are unchanged by this correction.