feat: end to end connected for demo

Includes the Transcription agent. Involved updating the RI agent to receive messages from other agents, sending speech commands to the RI agent, and some performance optimizations.

ref: N25B-216

src/control_backend/agents/vad_agent.py

@@ -55,8 +55,19 @@ class Streaming(CyclicBehaviour):
 
         self.audio_buffer = np.array([], dtype=np.float32)
         self.i_since_speech = 100  # Used to allow small pauses in speech
+        self._ready = False
+
+    async def reset(self):
+        """Clears the ZeroMQ queue and tells this behavior to start."""
+        discarded = 0
+        while await self.audio_in_poller.poll(1) is not None:
+            discarded += 1
+        logging.info(f"Discarded {discarded} audio packets before starting.")
+        self._ready = True
 
     async def run(self) -> None:
+        if not self._ready: return
+
         data = await self.audio_in_poller.poll()
         if data is None:
             if len(self.audio_buffer) > 0:
@@ -108,6 +119,8 @@ class VADAgent(Agent):
         self.audio_in_socket: azmq.Socket | None = None
         self.audio_out_socket: azmq.Socket | None = None
 
+        self.streaming_behaviour: Streaming | None = None
+
     async def stop(self):
         """
         Stop listening to audio, stop publishing audio, close sockets.
@@ -150,8 +163,8 @@ class VADAgent(Agent):
             return
         audio_out_address = f"tcp://localhost:{audio_out_port}"
 
-        streaming = Streaming(self.audio_in_socket, self.audio_out_socket)
-        self.add_behaviour(streaming)
+        self.streaming_behaviour = Streaming(self.audio_in_socket, self.audio_out_socket)
+        self.add_behaviour(self.streaming_behaviour)
 
         # Start agents dependent on the output audio fragments here
         transcriber = TranscriptionAgent(audio_out_address)