Tags: ashusevim/spellguard

Accuracy fixes: tech-word supplement, dot-segment tokens, repeated-wo…

…rd review

Dogfooding 'spellguard --agent README.md' produced 11 findings, most of
them garbage. Three deterministic fixes:

- techdict.ts: ~200-word supplement of common tech/prose words hunspell
  en_US lacks (config, msg, auth, roadmap, transpiler, retarget, utf,
  file extensions, spell-checker family). Matched case-insensitively.
  Kills the msg->mag / config->coning / UTF->but false positives.
- Dot-segmented tokens (Node.js, package.json, README.md) are valid when
  every dot-segment is a valid word; splitIdentifiers now splits on dots
  so teh.js flags 'teh', not 'tehjs'. Kills Nodejs->node's garbage.
- --agent: misspellings repeated across >= 2 lines downgrade to [review]
  ('appears on N lines - if intentional, bless') - repeated words are
  usually intentional vocabulary (aspell->ispell class), not typos.

Dogfood result: the same README.md run now reports NO ISSUES.

Tests: 104 passing (techdict 6, CLI +1, one fixture corrected: kubernets
-> recieve since kubernetes is not in hunspell either)

Rename package to spellguard

- package.json: name + bin key spellguard (verified available on npm)
- index.ts: CLI_NAME heuristic - agent task lists and help print
  'spellguard' when running as the installed bin, 'node index.ts' in dev
- README: install commands (npm i -g spellguard / npx spellguard)
- Verified: npm pack -> fresh consumer install -> npx spellguard
  --agent/--fix end to end; self-service commands show the bin name