feat: add PDF test corpus (1,905 files, 10 source groups)

[StephanMeijer]

Summary

Adds 1,905 openly-licensed PDFs as a new format to the DocSpec test corpus, across 10 diverse source groups.

Source Groups

Group	Files	License	Coverage
pdf-js	711	Apache-2.0	Annotations, forms, CID fonts, Arabic/CJK, encryption, edge cases
qpdf	649	Apache-2.0	Encryption (AES/RC4), page operations, compression, malformed PDFs
apache-pdfbox	154	Apache-2.0	Forms/AcroForms, text extraction, image handling
pikepdf	35	MPL-2.0	Color spaces, PDF/A compliance
arxiv	65	CC-BY-4.0	Academic papers, equations, multi-column, diverse disciplines
openstax	23	CC-BY-4.0	Textbook excerpts, tables, figures, equations
irs-forms	144	CC0-1.0	Fillable forms, checkboxes, government structured forms
nasa-ntrs	86	CC0-1.0	Technical reports, figures, tables, graphs
gutenberg	10	CC0-1.0	Public domain literature (EN, DE, FR, ES, IT)
internet-archive	28	CC0-1.0	Scanned/OCR books (Arabic, Chinese, French, German, Spanish, Russian, English)

Validation

• ✅ All 1,905 PDFs start with `%PDF-` magic bytes
• ✅ No file exceeds 10MB
• ✅ No duplicate files (`bin/check-duplicates.sh`)
• ✅ All SPDX license identifiers valid (`bin/validate-licenses.sh`)
• ✅ ATTRIBUTION.json updated with 10 entries (one per group)
• ✅ ATTRIBUTION.md regenerated
• ✅ All PDFs tracked via Git LFS (`.gitattributes` unchanged)
• ✅ No download scripts committed (`scrape/` is gitignored)

Feature Coverage

text-heavy · tables · forms · fillable-forms · images · equations · multilingual · rtl · cjk · scanned · ocr · annotations · encryption · test-fixtures · multi-column · public-domain · academic · government · educational · technical-reports