v4.10.5

DonSeTch v4.10.5

Fixed

• donsetch stop keeps live temporary browser profiles whose names include
  a nonce and sequence, as well as the older PID-only names. Cleanup had
  mistaken every current-format profile for a crash leftover (#397).
• PDF word spacing uses glyph bounds including side bearings rather than
  ink gaps, preserving numbers such as 15 % and exact must_contain
  matches. Ink bounds still drive layout and table geometry. Rotated PDF
  text also keeps its reading direction through frame canonicalization (#399).
• Fetch language metadata no longer identifies Latin text as English,
  Cyrillic text as Russian, or Devanagari text as Hindi. lang is present
  only for supported HTML declarations, with lang_source; observed writing
  scripts use a separate script field. PDFs and API cards no longer claim
  an unestablished language. Undeclared text retains script-aware tokenization
  and accent folding without language-specific stemming (#398).
• Google's browser search lane keeps one or two usable results instead of
  discarding them as a captcha. Zero parsed results report empty-parse;
  confirmed captcha failures retain blocked:captcha. The existing breadcrumb
  URL limitations are documented and retained in v4 (#396).

Changed

• Additions to the native featured BYOK search-provider list are paused.
  Existing providers remain supported; new integrations can use the existing
  search plugin interface.

What's Changed

• fix: preserve browser sessions, PDF text and language metadata by @dondai44423 in #401

Full Changelog: v4.10.4...v4.10.5