feat: improve PDF extraction for financial reports and garbled layouts

[ivanvanderbyl]

Summary

• improve financial-report extraction with region-aware table rendering, safer subtitle/note handling, and better preserved leading table columns
• add suspicious local block recovery for rotated and fragmented text, including recovered issue-140 table output and conservative garble cleanup
• add high-confidence encoding-noise suppression for issue-192 so structured anchors are preserved while low-confidence gibberish is dropped

Test Plan

• go test ./... -count=1
• go run ./cmd/pdfmarkdown --input '/Users/ivanvanderbyl/dev/Alcova-AI/pdf-markdown/testdata/issue-140-example.pdf'
• go run ./cmd/pdfmarkdown --input '/Users/ivanvanderbyl/dev/Alcova-AI/pdf-markdown/testdata/issue-192-example.pdf'

[ivanvanderbyl]

This change is part of the following stack:

• feat: improve PDF extraction for financial reports and garbled layouts #4 ◀

_{Change managed by git-spice.}