feat(tts): add Doubao TTS 2.0 (Volcengine Seed-TTS 2.0) provider

[wyuc]

Summary

• Add Doubao TTS 2.0 (Volcengine Seed-TTS 2.0) as a new TTS provider with 17 voices, streaming response parsing, and rate-limit error handling
• Full integration across types, constants, provider implementation, server config, settings store, i18n (zh-CN + en-US), and UI components
• Follows the existing TTS provider pattern - no architectural changes

Closes #282

Changed files

File	Change
lib/audio/types.ts	Add doubao-tts to TTSProviderId union
lib/audio/constants.ts	Provider config (17 voices) + default voice
lib/audio/tts-providers.ts	TTSRateLimitError + generateDoubaoTTS()
lib/server/provider-config.ts	TTS_DOUBAO env var mapping
lib/store/settings.ts	doubao-tts default config entry
lib/i18n/settings.ts	zh-CN and en-US translations
3 UI components	doubao-tts in provider name maps

Test plan

• pnpm check / pnpm lint / npx tsc --noEmit pass
• Configure Doubao TTS in Settings > Audio with appId:accessKey and verify voice preview works
• Verify existing TTS providers are unaffected

🤖 Generated with Claude Code

[cosarah]

Issues

Important

• lib/audio/tts-providers.ts:109 — TTSRateLimitError is exported and thrown but never caught distinctly from Error anywhere. If retry/backoff logic is planned, add a note or TODO; otherwise this is dead code.

• lib/audio/tts-providers.ts:484 — atob() + manual byte copy is less robust and less efficient than Buffer.from(chunk.data, 'base64') for server-side code. Consider:

  const bytes = Buffer.from(chunk.data, 'base64');
  audioChunks.push(new Uint8Array(bytes));

• tts-settings.tsx:121,149 — "App ID" and "Access Key" labels are hard-coded English strings. Per project convention, UI text needs i18n — use t('settings.doubaoAppId') / t('settings.doubaoAccessKey') with entries in lib/i18n/settings.ts.

• lib/audio/tts-providers.ts:452 + constants.ts — supportedFormats declares ['mp3', 'ogg_opus', 'pcm'] but the request always hard-codes format: 'mp3'. Either use the format from config or trim supportedFormats to ['mp3'] to avoid misleading callers.

Suggestions

• getTTSProviderName is duplicated across 3 files (pre-existing, but this PR deepens it). Worth extracting to a shared utility.
• Access Key input field lacks the show/hide toggle that App ID has — minor inconsistency.

Summary

Well-structured PR that follows existing TTS provider patterns. The i18n gap and atob → Buffer.from fix should be addressed before merge; the rest are minor.

[cosarah]

Previous review items are all properly addressed in aa6b28b ✓

• TTSRateLimitError: TODO comment added — acceptable
• atob() → Buffer.from(): fixed
• i18n labels: fixed, both locales updated
• supportedFormats: trimmed to ['mp3']
• Bonus: Access Key show/hide toggle added

No new blocking issues. The duplicated getTTSProviderName across 3 files is pre-existing debt worth a follow-up issue.

LGTM — ready to merge.