feat: Add Speech-to-Text voice input for message composer

[Makisuo]

Summary

Allow users to dictate messages using voice input, with speech automatically transcribed to text in the message composer.

Feature Description

Users should be able to:

• Click a microphone button in the message composer to start voice recording
• Speak their message and have it transcribed to text in real-time
• See visual feedback while recording (e.g., pulsing indicator, recording state)
• Cancel or confirm the transcription before sending

UI Location

• Add a microphone button in the message composer actions (alongside attach and emoji buttons)
• Show recording state indicator when active
• Display transcription in the Slate editor as it's processed

Suggested Implementation

1. Create a useSpeechRecognition hook using the Web Speech API (SpeechRecognition)
2. Add microphone button to message-composer-actions.tsx
3. Handle browser compatibility (Web Speech API is primarily Chrome/Edge/Safari)
4. Provide fallback or error message for unsupported browsers
5. Request microphone permissions and handle denial gracefully

Technical Considerations

• Web Speech API (browser-native) vs external API (e.g., OpenAI Whisper)
• Push-to-talk vs toggle recording UX
• Accessibility: ensure voice input doesn't exclude keyboard-only users