test: complete VAD unit and integration tests

Including an integration test with real voice audio.

ref: N25B-213

test/integration/agents/vad_agent/test_vad_with_audio.py

@@ -0,0 +1,57 @@
+import os
+from unittest.mock import MagicMock, AsyncMock
+
+import pytest
+import soundfile as sf
+import zmq
+
+from control_backend.agents.vad_agent import Streaming
+
+
+def get_audio_chunks() -> list[bytes]:
+    curr_file = os.path.realpath(__file__)
+    curr_dir = os.path.dirname(curr_file)
+    file = f"{curr_dir}/speech_with_pauses_16k_1c_float32.wav"
+
+    chunk_size = 512
+
+    chunks = []
+
+    with sf.SoundFile(file, 'r') as f:
+        assert f.samplerate == 16000
+        assert f.channels == 1
+        assert f.subtype == "FLOAT"
+
+        while True:
+            data = f.read(chunk_size, dtype="float32")
+            if len(data) != chunk_size:
+                break
+
+            chunks.append(data.tobytes())
+
+    return chunks
+
+
+@pytest.mark.asyncio
+async def test_real_audio(mocker):
+    """
+    Test the VAD agent with only input and output mocked. Using the real model, using real audio as
+    input. Ensure that it outputs some fragments with audio.
+    """
+    audio_chunks = get_audio_chunks()
+    audio_in_socket = AsyncMock()
+    audio_in_socket.recv.side_effect = audio_chunks
+
+    mock_poller: MagicMock = mocker.patch("control_backend.agents.vad_agent.zmq.Poller")
+    mock_poller.return_value.poll.return_value = [(audio_in_socket, zmq.POLLIN)]
+
+    audio_out_socket = AsyncMock()
+
+    vad_streamer = Streaming(audio_in_socket, audio_out_socket)
+    for _ in audio_chunks:
+        await vad_streamer.run()
+
+    audio_out_socket.send.assert_called()
+    for args in audio_out_socket.send.call_args_list:
+        assert isinstance(args[0][0], bytes)
+        assert len(args[0][0]) >= 512*4*3  # Should be at least 3 chunks of audio