Add CAPI support for MAI in voice mode and dictation

Add CAPI support for MAI so it can be used by both voice mode and dictation.

## Scope

- Enable MAI through CAPI for voice mode.
- Enable MAI through CAPI for dictation.
- Ensure both experiences use the appropriate CAPI request/response flow and handle errors consistently with existing providers.
- Add coverage for provider selection, successful transcription, and failure/cancellation paths.

## Done when

- Voice mode can use MAI through CAPI end to end.
- Dictation can use MAI through CAPI end to end.
- Existing voice mode and dictation providers continue to work unchanged.
- Relevant automated tests and documentation are updated.