Use dedicated MAI transcription for dictation

[Arthur Cnops (accnops)]

MAI dictation now uses the dedicated /api/v1/realtime/transcription WebSocket instead of sharing Voice Mode's /voice session.

• derives the transcription endpoint without changing Voice Mode routing
• adds a strict, lifecycle-safe transcription transport with ping/pong and terminal error handling
• treats MAI final transcripts as authoritative, including shorter or empty finals
• waits for the backend's bounded finalization window while allowing cancel and rapid restart to finish immediately
• preserves the existing local dictation path

Fixes #332324.

This restores and adapts the dedicated transcription-client direction from #326228 by Mir Imad Ahmed.

Testing:

• focused browser tests: 70 passed
• touched-file TypeScript diagnostics: none
• ESLint on all changed files
• hygiene pre-commit hook
• Fallow new-only audit: pass

[Copilot]

🟡 Changes recommended

Voice routing changes unexpectedly, and stale socket errors can terminate a restarted dictation session.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Moves MAI dictation from Voice Mode’s shared socket to a dedicated transcription transport.

Changes:

• Adds endpoint derivation and a lifecycle-aware transcription client.
• Integrates authoritative MAI finals and bounded finalization.
• Adds focused endpoint, transport, and dictation tests.

File summaries

File	Description
voiceEndpoint.ts	Resolves voice/transcription URLs and authentication.
voiceClientService.ts	Uses shared endpoint helpers.
voiceCodeTranscriptionClient.ts	Implements dedicated WebSocket transport.
chatSpeechToTextService.ts	Integrates MAI transcription and finalization.
chat.shared.contribution.ts	Registers the transcription service.
voiceEndpoint.test.ts	Tests endpoint handling.
voiceCodeTranscriptionClient.test.ts	Tests transport lifecycle and protocol.
chatSpeechToTextService.test.ts	Tests MAI integration and final selection.

Review details

• Files reviewed: 8/8 changed files
• Comments generated: 4
• Review effort level: Balanced

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.