refactor(backend): 启动环境初始化与资源项目化（零跨项目依赖）

- main.py：uv run main.py 一键启动——读取 backend/.env、TORCH_HOME/NLTK_DATA 指向项目内缓存、certifi SSL + NO_PROXY 网络直连、默认开启 s2s，启动横幅
- vendor/：补丁后 s2s 云化栈移入项目（backend/vendor/s2s-cloud），pyproject sources 改相对路径
- .gitignore：忽略 .torch-cache / nltk_data 缓存
- AGENTS.md：进程管理铁律（AI 不启动/重启服务，由用户操作）

backend/vendor/s2s-cloud/tests/test_websocket_session_lifecycle.py

@@ -0,0 +1,94 @@
+import asyncio
+from queue import Queue
+from threading import Event, Thread
+
+from speech_to_speech.baseHandler import BaseHandler
+from speech_to_speech.connections.websocket_streamer import WebSocketStreamer
+from speech_to_speech.pipeline.control import SESSION_END, is_control_message
+from speech_to_speech.pipeline.messages import PIPELINE_END
+
+
+class EchoHandler(BaseHandler):
+    def setup(self):
+        self.processed = []
+        self.session_end_calls = 0
+
+    def process(self, item):
+        self.processed.append(item)
+        yield item.upper()
+
+    def on_session_end(self):
+        self.session_end_calls += 1
+
+
+class FakeWebSocket:
+    def __init__(self, messages):
+        self._messages = iter(messages)
+
+    def __aiter__(self):
+        return self
+
+    async def __anext__(self):
+        try:
+            return next(self._messages)
+        except StopIteration as exc:
+            raise StopAsyncIteration from exc
+
+
+def test_base_handler_session_end_resets_without_stopping():
+    stop_event = Event()
+    queue_in = Queue()
+    queue_out = Queue()
+    handler = EchoHandler(stop_event, queue_in=queue_in, queue_out=queue_out)
+
+    thread = Thread(target=handler.run)
+    thread.start()
+
+    queue_in.put(SESSION_END)
+    queue_in.put("hello")
+    queue_in.put(PIPELINE_END)
+
+    thread.join(timeout=2)
+    assert not thread.is_alive()
+
+    outputs = [queue_out.get(timeout=1) for _ in range(3)]
+    assert is_control_message(outputs[0], SESSION_END.kind)
+    assert outputs[1] == "HELLO"
+    assert outputs[2] == PIPELINE_END
+    assert handler.processed == ["hello"]
+    assert handler.session_end_calls == 1
+
+
+def test_websocket_streamer_last_disconnect_queues_session_end():
+    streamer = WebSocketStreamer(
+        stop_event=Event(),
+        input_queue=Queue(),
+        output_queue=Queue(),
+        should_listen=Event(),
+    )
+
+    asyncio.run(streamer._handle_client(FakeWebSocket([])))
+
+    queued = streamer.input_queue.get_nowait()
+    assert is_control_message(queued, SESSION_END.kind)
+    assert streamer.should_listen.is_set()
+
+
+def test_websocket_send_loop_ignores_session_end_until_stop():
+    stop_event = Event()
+    streamer = WebSocketStreamer(
+        stop_event=stop_event,
+        input_queue=Queue(),
+        output_queue=Queue(),
+        should_listen=Event(),
+    )
+
+    async def exercise_send_loop():
+        task = asyncio.create_task(streamer._send_loop())
+        streamer.output_queue.put(SESSION_END)
+        await asyncio.sleep(0.05)
+        assert not task.done()
+        stop_event.set()
+        await asyncio.wait_for(task, timeout=1)
+
+    asyncio.run(exercise_send_loop())