- PDF pages rendered via PDF.js (see actual document, not just extracted text)
- Text/PDF view toggle appears for PDF documents (top-right of header)
- PDF view scrolls to current page as sentences are read
- Column-aware extraction: detects two-column layouts, reads left column
top-to-bottom then right column (fixes academic papers, double-page PDFs)
- De-hyphenation: removes mid-word line breaks from extracted text
- Both PDF upload and Nextcloud download now return pdf_b64 + page_texts
- sentence→page mapping so PDF view auto-scrolls during playback
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
00af54c95a
