Per-session orchestrator: inbound audio → STT → turn callback → TTS → outbound audio.

Decoupled from any specific app. The host provides:

• :turn_fn — fn text -> {:ok, response} | {:error, reason} end

• :endpoint — Phoenix Endpoint module for broadcasting to channels
• :pubsub — Phoenix.PubSub module for internal notifications
• :config — VoiceEngine.Config struct with STT/TTS adapters + opts

Starting

VoiceEngine.Bridge.start_link(
  session_id: "abc-123",
  config: VoiceEngine.Config.current(:my_app),
  turn_fn: &MyApp.ConvEngine.turn(session_id, &1),
  endpoint: MyAppWeb.Endpoint,
  pubsub: MyApp.PubSub
)