0.15.0

Cascade gets the Trafilatura ratio condition and an empty-URL bug
fix; anchor election arrives as a structural inversion (StripOptions::main_landmark); a committed regression corpus pins per-fixture F1 floors so future filter changes that drop F1 fail CI.

Highlights

• extract_html_cascade now uses Trafilatura's 2x ratio
  condition (the only published cascade form with measured F1
  evidence) instead of the prior fixed-threshold form. Short-circuits
  readability when the scanner already produced enough.
• Anchor election via StripOptions::main_landmark(): pre-scans
  for <main> or longest <article> and restricts extraction to
  that subtree. WCXB article F1 0.855 → 0.867. Opt-in (regresses
  forum / product pages where content lives outside <main>).
• Regression corpus (tests/fixtures/regression/) with per-
  extractor F1 floors. Failing CI on F1 drift, not just test counts.
• Bug fix: extract_with_readability("...", "") silently
  returned None because dom_smoothie rejected the empty URL. The
  cascade's fallback branch had been disabled in practice; now
  fires correctly on real-world content-in-aside pages.

See CHANGELOG for the full list.

Test counts

677 tests + 17 doc tests passing under --all-features (was 661).