Add Phrases: trigger → snippet substitution for dictated text

[mattmoran56]

Summary

Adds a new Phrases section to the Control Panel sidebar where users define trigger → snippet pairs. After Whisper transcribes their speech, each trigger phrase is replaced with the corresponding snippet while surrounding text is preserved verbatim.

Use cases: dictate "my email" and have it expand to an email address; dictate "message template one best wishes Matt" and have the template expand while "best wishes Matt" stays intact.

Changes

• New Phrases view (src/components/PhrasesView.tsx) with empty state, inline-editable trigger / snippet rows, auto-growing snippet textarea (CSS field-sizing: content), and add/delete controls
• Sidebar entry wired through ControlPanelSidebar.tsx and ControlPanel.tsx (new phrases view, Replace icon)
• Substitution engine (src/utils/phraseSubstitution.js) — case-insensitive, Unicode-aware word boundaries (\p{L}\p{N}), longer triggers win when they overlap, regex metacharacters escaped, blank triggers skipped
• Single pipeline injection in audioManager.processAudio after reasoning, before paste/save — covers every transcription provider (local Whisper, Parakeet, OpenWhispr cloud, OpenAI). Raw transcript is preserved in DB; only the substituted text reaches the clipboard
• Persistence via customPhrases in settingsStore (localStorage, mirroring customDictionary); typed CustomPhrase shape in src/types/phrases.ts
• i18n — sidebar.phrases + phrases.* keys translated into all 10 locales (en source + de, es, fr, it, ja, pt, ru, zh-CN, zh-TW). Each non-English phrases block carries a _comment field noting the translations were AI-generated by Claude and have not been reviewed by a native speaker
• Tests — test/helpers/phraseSubstitution.test.js (12 cases) using node:test, matching the repo's existing test convention. Adds an npm test script that runs the full test/**/*.test.js suite (24 tests total alongside the 12 pre-existing helper tests)

Out of scope (follow-up)

• Streaming dictation bypasses processAudio; phrase substitution only runs on the post-transcription path for v1
• Snippet templating ({{date}} etc.) is not supported

Testing

• npm run lint — 0 errors (3 pre-existing warnings unrelated to this change)
• npm run typecheck — clean
• npm test — 24/24 pass (12 new + 12 pre-existing). New cases: basic substitution, trailing punctuation preserved, case-insensitive matching, plural word-boundary non-match, longest-trigger-wins, empty-list no-op, leading-position match, multi-line snippets, regex-metachar escaping, blank-trigger handling, multiple occurrences, empty text
• node scripts/check-i18n.js — locale keys and placeholders consistent across all 10 locales
• Manual: open Control Panel → new Phrases sidebar entry renders, empty state shows "Add your first phrase" CTA
• Manual: add a phrase my email → me@example.com, verify it persists across reloads (localStorage)
• Manual: dictate "send my email please" — clipboard receives "send me@example.com please" (trigger expanded, surrounding text preserved)
• Manual: dictate "message template one, best wishes Matt" with message template one → multi-line template — trailing ", best wishes Matt" preserved verbatim after the inserted template
• Manual: dictate with no matching phrase — output unchanged (no-op when triggers don't fire)
• Manual: dictate "my emails are slow" with trigger my email — does not match (word-boundary respect verified end-to-end)
• Manual: snippet textarea auto-grows vertically as content is typed
• Manual: deleting a phrase removes it from subsequent dictations
• Manual: History view still shows the raw spoken words (rawText preserved in DB), while clipboard/preview show the substituted text

---

📝 Personal note: this is one of my favourite features on the paid Whispr apps — it'd be awesome to see it land here. Happy to revise direction, scope, or naming based on feedback!