fix: allow Whisper to generate more tokens based on audio length

Before, it sometimes cut off the transcription too early.

ref: N25B-209

src/control_backend/agents/transcription/transcription_agent.py

@@ -59,6 +59,10 @@ class TranscriptionAgent(Agent):
             audio = await self.audio_in_socket.recv()
             audio = np.frombuffer(audio, dtype=np.float32)
             speech = await self._transcribe(audio)
+            if not speech:
+                logger.info("Nothing transcribed.")
+                return
+
             logger.info("Transcribed speech: %s", speech)
 
             await self._share_transcription(speech)