cli-v1.7.2

Added

• [EXPERIMENTAL] OCR for scanned PDFs (pdf-ocr build feature) — epublift import now converts pure scans (PDFs with no text layer) using a 100%
  pure-Rust OCR pipeline (ocrs on the rten tensor engine), with flat-field
  illumination correction for phone-photo scans. The ~12 MB models download on
  first use and cache (override with EPUBLIFT_OCR_MODELS); the downloader uses
  the same pure-Rust rustls/RustCrypto client as meta enrich (no C). OCR is
  best-effort: ~92% word accuracy on a clean prose scan, lower on
  skewed/low-contrast/photographed pages — it gets readable reflowable text out
  of an otherwise-unreadable image PDF, not a faithful reproduction. The web
  service does not OCR yet (long jobs need async handling); scanned uploads
  are still reported there. See docs/pdf-import.md.

Full Changelog: cli-v1.7.1...cli-v1.7.2