ePubLift/epublift cli-v5.1.0 on GitHub

import reads PDFs through paginaverto 0.3. What a reader notices:
- OCR reads Turkish and the other Latin-script alphabets. Recognition now
  uses PaddlePaddle's PP-OCRv5 Latin model, so ş ğ ı İ ç ö ü come through.
  On a 292-page Turkish scan, 88.4% of the words came out exactly right, in
  order, against the publisher's EPUB; the old model got 47.1% on the same
  pages. With --language tr, İ and ı are told from I and i (983
  of 1,059 capital İ right). The first scan downloads 8.0 MB more.
- Scans in JBIG2 and JPEG 2000 are read instead of refused (that Turkish
  scan was one), and figures in those formats and CCITT reach the book.
  Covers, plates and end papers of a scan with a text layer are kept.
- Chapters follow the PDF's own outline when it is a table of contents,
  else a printed contents page. Headings inside a chapter become h2 to
  h6, and italic text stays italic.
- Pages that could not be read are named in a warning, and the rest of
  the book is written.
- The OCR warning is only for languages the model cannot read: a scan
  in a non-Latin script (Cyrillic, Greek, Arabic, CJK and others), or in
  Azerbaijani, Vietnamese or Uzbek, whose letters it lacks, says the text is
  wrong or missing.
- The book's title comes from the PDF (its document properties) when it
  has a real one, else from the file name as before.
  docs/pdf-import.md has the numbers.

Full Changelog: cli-v5.0.0...cli-v5.1.0