Modernize to TypeScript + ESM CLI; drop lex/yargs; republish

[mvhenten]

Revive and modernize this i18n string extractor. It was unpublished from npm (2025-06-19); republish after modernizing. Small, coherent tool (~100 lines) with a CLI, and existing tests guard the tricky lexer.

Why

• CommonJS, untyped, and leans on the unmaintained lex lexer + yargs.
• Core logic is simple enough to own outright with near-zero deps.

Tasks

• Convert to src/ TypeScript (strict). Preserve behavior exactly (tests lex_test/scan_test are the guard):
  • tokenize string literals inside __('...') / __n('...') calls, incl. escaped-quote handling
  • scan files → unique, sorted token list
• Drop lex → inline a small (~30-line) tokenizer/state-machine.
• Drop yargs → node:util parseArgs. Consider dropping debug too.
• Wire the bin (extract-gettext) through package.json; ensure the CLI runs under ESM.
• Port tape tests → node:test.
• CI (Node 20/22/24) + release workflow with npm trusted publishing (--provenance). Mirror mvhenten/phishy.
• Republish once trusted publishing is enabled.

Notes

• Module format: ESM-only is fine for the CLI; offer a dual lib build only if library (non-CLI) consumers matter.
• Template/reference: mvhenten/phishy.
• Publishing setup to be enabled by @mvhenten before release.
• No AI attribution on commits.