fromGeminiLive should expose output transcription (assistant spoken text)

[octalpixel]

What

fromGeminiLive's options don't expose output transcription (enableOutputTranscription / outputAudioTranscription). Input transcription is hardcoded on (inputAudioTranscription: {}) and output transcription isn't configurable at all.

This blocks surfacing the assistant's spoken text to the client — the other half of a two-sided transcript UI (the app can render the user's turns but has no assistant transcript to display). Related to #28 (input transcription) and #29 (options-surface passthrough).

Acceptance criteria

• fromGeminiLive({ ... }) can enable/receive output (assistant) transcription, surfaced as a client-consumable event.
• (Consistency) input transcription is opt-in/configurable too, rather than hardcoded.

Blocked by

None — can start immediately.