🔩 Improve footnote Markdown copying and require reviewed OCR

[JFWooten4]

Summary

This branch changes two connected parts of the PDF reading/enhancement workflow:

• More faithful footnotes in copied Markdown. Render footnote references as inline Unicode superscript numbers, place each corresponding [number]: definition directly after the paragraph containing its citation, and preserve bold/italic formatting from the note's PDF text spans. Automatically include an indexed footnote even when the copied selection contains its citation but not the full footnote text.
• Full image-based OCR review for browser enhancement. Make both native-host light and deep enhancement modes pass --force-ocr --ai-review instead of --skip-text, so even PDFs with an existing (possibly inaccurate) OCR layer are re-read and locally reviewed against page images.

Details

Footnote selection and formatting

• Factor paragraph boundary handling into startsParagraph and apply it consistently when rendering text and reference markers.
• Recover full footnote styling from text-layer spans, including unselected portions of a note; use the indexed, OCR-corrected note text for the definition.
• Associate footnote definitions with the citing paragraph rather than appending all notes after the copied passage; include separately selected notes once without duplicating definitions.
• Match a reference to its actual highlight region rather than requiring the note's starting pageNumber to equal the reference's page, accommodating indexed regions extending across pages.
• Preserve the existing plain-text/native-copy guard and page/paragraph boundary behavior.

This builds on the copied-footnote recovery introduced by #183 and its subsequent same-page/page-boundary correction in #184.

Enhancement pipeline

• Replace the text-preserving browser behavior established in 🔧 Simplify browser PDF enhancement and version cached results #140 and retained by 🔧 Preserve source links and text during PDF enhancement #141 with --force-ocr --ai-review for both browser modes.
• Increment PIPELINE_VERSION from 3 to 4 so earlier completed enhancements are regenerated instead of reusing outputs made under the old pipeline.
• Update the README to explain that the enhanced copy is rasterized, can take substantially longer, and does not modify the original source PDF.

Regression coverage

• Extend the real SEC petition fixture to cover footnotes 11 and 12, superscript markers, inline definition placement, partially selected paragraphs, selected and unselected note text, duplicate avoidance, and bold/italic formatting.
• Update native host and CLI tests for the new OCR arguments and mode mapping.

The added tests are present in the branch, but I have not executed them locally in this workflow.

Review considerations

The OCR default is a deliberate cost and fidelity tradeoff: browser enhancement now re-rasterizes every page and runs local AI review, even when an embedded text layer is present. Reviewers should verify runtime/performance and any downstream implications for selectable text and accessibility.