feat(adapters): 实现三层流式看门狗

TTFT / inter_token / total 超时保护。
参考 CHSAnalyzer2 streaming.py。

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

tests/unit/test_streaming.py

@@ -0,0 +1,152 @@
+"""三层流式看门狗 (stream_with_liveness_timeouts) 单元测试。
+
+验证 TTFT / inter_token / total 三层超时保护行为，
+以及正常透传、空流等边界场景。
+"""
+
+from __future__ import annotations
+
+import asyncio
+from typing import TYPE_CHECKING, TypeVar
+
+if TYPE_CHECKING:
+    from collections.abc import AsyncIterator
+
+import pytest
+
+from adapters.streaming import StreamLivenessTimeout, stream_with_liveness_timeouts
+
+_T = TypeVar("_T")
+
+
+async def _items_with_delays(
+    items: list[_T],
+    delays: list[float],
+) -> AsyncIterator[_T]:
+    """辅助异步生成器：按 delays 指定的延迟逐项产出 items。
+
+    参数:
+        items: 待产出的元素列表。
+        delays: 每个元素产出前的延迟秒数，长度须与 items 一致。
+    """
+    for item, delay in zip(items, delays, strict=True):
+        await asyncio.sleep(delay)
+        yield item
+
+
+@pytest.mark.asyncio
+async def test_all_items_yielded_within_budget() -> None:
+    """正常场景：所有 item 在超时预算内产出，全部透传。"""
+    items = ["a", "b", "c", "d"]
+    delays = [0.01, 0.01, 0.01, 0.01]
+
+    result: list[str] = []
+    async for chunk in stream_with_liveness_timeouts(
+        _items_with_delays(items, delays),
+        ttft_s=1.0,
+        inter_token_s=1.0,
+        total_s=5.0,
+    ):
+        result.append(chunk)
+
+    assert result == items
+
+
+@pytest.mark.asyncio
+async def test_ttft_timeout_fires() -> None:
+    """首 token 超时：第一个 item 延迟超过 ttft_s，触发 kind='ttft'。"""
+    items = ["slow_first"]
+    delays = [0.5]
+
+    with pytest.raises(StreamLivenessTimeout) as exc_info:
+        async for _ in stream_with_liveness_timeouts(
+            _items_with_delays(items, delays),
+            ttft_s=0.05,
+            inter_token_s=5.0,
+            total_s=10.0,
+        ):
+            pass  # pragma: no cover — 不应到达
+
+    err = exc_info.value
+    assert err.kind == "ttft"
+    assert err.first_token_seen is False
+
+
+@pytest.mark.asyncio
+async def test_inter_token_timeout_fires() -> None:
+    """token 间隔超时：首项正常，第二项延迟超过 inter_token_s。"""
+    items = ["fast", "very_slow"]
+    delays = [0.01, 0.5]
+
+    with pytest.raises(StreamLivenessTimeout) as exc_info:
+        async for _ in stream_with_liveness_timeouts(
+            _items_with_delays(items, delays),
+            ttft_s=5.0,
+            inter_token_s=0.05,
+            total_s=10.0,
+        ):
+            pass
+
+    err = exc_info.value
+    assert err.kind == "inter_token"
+    assert err.first_token_seen is True
+
+
+@pytest.mark.asyncio
+async def test_total_timeout_fires() -> None:
+    """总时长超时：每个 item 都在单层预算内，但累计超过 total_s。"""
+    items = ["a", "b", "c", "d", "e", "f"]
+    # 每项 0.05s，累计 0.30s > total_s=0.15s
+    delays = [0.05, 0.05, 0.05, 0.05, 0.05, 0.05]
+
+    with pytest.raises(StreamLivenessTimeout) as exc_info:
+        async for _ in stream_with_liveness_timeouts(
+            _items_with_delays(items, delays),
+            ttft_s=5.0,
+            inter_token_s=5.0,
+            total_s=0.15,
+        ):
+            pass
+
+    err = exc_info.value
+    assert err.kind == "total"
+
+
+@pytest.mark.asyncio
+async def test_thinking_tokens_refresh_watchdog() -> None:
+    """(False, 'think') 元组正常透传，不影响看门狗计时。"""
+    items = [
+        (False, "think_1"),
+        (False, "think_2"),
+        (True, "real_token"),
+    ]
+    delays = [0.01, 0.01, 0.01]
+
+    result: list[tuple[bool, str]] = []
+    async for chunk in stream_with_liveness_timeouts(
+        _items_with_delays(items, delays),
+        ttft_s=1.0,
+        inter_token_s=1.0,
+        total_s=5.0,
+    ):
+        result.append(chunk)
+
+    assert result == items
+
+
+@pytest.mark.asyncio
+async def test_empty_stream_no_error() -> None:
+    """空流：source 立即耗尽，不抛异常，正常结束。"""
+    items: list[str] = []
+    delays: list[float] = []
+
+    result: list[str] = []
+    async for chunk in stream_with_liveness_timeouts(
+        _items_with_delays(items, delays),
+        ttft_s=1.0,
+        inter_token_s=1.0,
+        total_s=5.0,
+    ):
+        result.append(chunk)  # pragma: no cover
+
+    assert result == []