feat: allow no audio input while robot is speaking

The VAD agent will discard its current buffer and retry receiving data.

ref: N25B-213

test/unit/agents/test_vad_streaming.py

@@ -86,9 +86,7 @@ async def test_no_data(audio_in_socket, audio_out_socket, streaming):
     audio_in_poller.poll.return_value = None
     streaming.audio_in_poller = audio_in_poller
 
-    assert streaming.i_since_data == 0
-
     await streaming.run()
 
     audio_out_socket.send.assert_not_called()
-    assert streaming.i_since_data == 1
+    assert len(streaming.audio_buffer) == 0