[低延迟端点]：完成首句保留和快速结束修正，包含ACK缓冲策略、批量读帧和主说话人端点回归测试

tests/test_transport.py

@@ -82,6 +82,15 @@ class TransportTests(unittest.TestCase):
         self.assertEqual(frames[0].pcm, b"\x01\x00\x02\x00")
         self.assertEqual(frames[0].sample_rate, 16000)
 
+    def test_sounddevice_transport_returns_queued_frame_batch(self) -> None:
+        fake = FakeSoundDevice(callback_payloads=[b"\x01\x00", b"\x02\x00", b"\x03\x00"])
+        transport = SoundDeviceAudioTransport(sounddevice_module=fake)
+        transport.start_input(sample_rate=16000, channels=1)
+        frames = transport.read_frames(10)
+        transport.stop()
+
+        self.assertEqual([item.frame_id for item in frames], [0, 1, 2])
+
     def test_sounddevice_transport_flushes_queued_input(self) -> None:
         fake = FakeSoundDevice()
         transport = SoundDeviceAudioTransport(sounddevice_module=fake)
@@ -115,7 +124,9 @@ class FakeInputStream:
 
     def start(self) -> None:
         self.started = True
-        self.callback(b"\x01\x00\x02\x00", 2, None, "")
+        payloads = self.kwargs.get("callback_payloads") or [b"\x01\x00\x02\x00"]
+        for payload in payloads:
+            self.callback(payload, max(1, len(payload) // 2), None, "")
 
     def stop(self) -> None:
         self.started = False
@@ -140,10 +151,13 @@ class FakeOutputStream:
 
 
 class FakeSoundDevice:
-    def __init__(self) -> None:
+    def __init__(self, callback_payloads: list[bytes] | None = None) -> None:
+        self.callback_payloads = callback_payloads
         self.output_writes: list[bytes] = []
 
     def RawInputStream(self, **kwargs):
+        if self.callback_payloads is not None:
+            kwargs["callback_payloads"] = self.callback_payloads
         return FakeInputStream(**kwargs)
 
     def RawOutputStream(self, **kwargs):