betto_lang_detector

Added example/example.dart and README.md (Features/Getting started/Usage/
Algorithm), recorded the manual accuracy spot-check findings in
docs/spec/README.md's Known limitations, and closed a coverage gap on
guess.dart's toString/==/hashCode overrides with a small dedicated test file.

make pre_commit is green (format, analyze, license_check via the real
addlicense binary, 74 tests) and make coverage reports 100% line coverage
(128/128) across every hand-written source file. dart pub publish --dry-run
reports 0 warnings and a correctly-scoped 67 KB archive.

Plan moved to docs/plans/completed/ with status Complete and a full summary.
Not pushed to GitHub, not published to pub.dev -- both are for the user to
do, per the agreed hand-off boundary.

Co-Authored-By: Claude Sonnet 5 [email protected]

Initial development release of betto_lang_detector.

Features

• LanguageDetector.pureDart() — zero-dependency default: a Unicode script pre-filter plus a character n-gram model, covering 58 languages (matching betto_lexical's Stopwords set).
• LanguageDetector.detect(String text) — returns Detected (a best guess plus a ranked list) or Undetermined (nothing met the confidence threshold).
• LanguageDetector.dominantScript(String text) — cheap, script-only classification (ISO 15924 code) for tokenizer routing guard rails; does not run the n-gram stage.
• restrictTo — narrows the candidate language set, improving n-gram accuracy for callers who already know the plausible language set.
• Pure Dart, zero runtime dependencies — no FFI, no native build, works identically on web, mobile, and desktop.