feat(chat): add voice input to message composer with dual-mode recognition

[louwetmarc657-cmd]

Summary

Adds a native voice input button (ComposerVoiceButton) to the chat message composer, enabling streaming speech-to-text input directly into the editor.

Key Features

• Dual-Mode Recognition:
  • Primary: Standard Web Speech API (webkitSpeechRecognition / SpeechRecognition) for instant, zero-latency streaming transcription.
  • Fallback / API: Supports media recording with OpenAI-compatible Whisper ASR endpoint when Web Speech is unavailable.
• User Experience:
  • Subtle waveform pulse feedback when recording is active.
  • Floating interim transcription preview bubble with keyboard escape (Esc) cancellation.
  • Seamless text insertion synchronized with Tiptap editor cursor and state.
• Internationalization:
  • Added localization keys in en.json and zh-CN.json.
• Tests:
  • Includes unit tests covering speech recognition lifecycles, permissions, state transitions, and UI interactions (composer-voice-button.test.tsx, use-voice-input.test.ts).

---

由社区用户贡献，遵循项目代码风格与规范。