feat: carry the per-call tier down to the transport that must send it

The Transport port took the request apart into five arguments, so a tier
placed on ChatRequest could never reach _build_payload: the field was set,
read by nobody, and silently ignored - the exact shape of failure that sent
downstream to extra_body in the first place.

complete() now takes reasoning_effort with no default, matching the
TelemetryRecorder convention: a default would turn a missing hand-off into
a silent 'no opinion'. All four fakes move with it, since @runtime_checkable
checks method names and not signatures.

EmbeddingTransport and OcrTransport are deliberately left alone - they have
no reasoning semantics - and a test now holds that line.

_build_payload drops its inline sugar conversion for effective_effort(), so
the guard and the hot path share one judgement, and passes the source's
effort_fallback for the same reason.

tests/unit/test_retry.py

@@ -24,6 +24,7 @@ from polygateway.types import (
     BackpressurePolicy,
     BreakerConfig,
     ChatRequest,
+    Effort,
     GlobalLimits,
     RetryPolicy,
     SourceConfig,
@@ -63,14 +64,21 @@ def _ok(content="ok"):
 
 
 class FakeTransport:
-    """按脚本逐次返回结果或抛异常;记录每次 (source_name, call_id)。"""
+    """按脚本逐次返回结果或抛异常;记录每次 (source_name, call_id) 与收到的档位。
+
+    `reasoning_effort` 刻意**不给默认值**,与 `Transport` 协议保持逐字一致:
+    `@runtime_checkable` 只查方法名不查签名,fake 上多一个默认值就会把"中间件漏传"
+    这类缺口伪装成"调用方没表态",而报错现场离根因很远。
+    """
 
     def __init__(self, script):
         self.script = list(script)
         self.calls = []
+        self.efforts = []
 
-    async def complete(self, *, messages, source, stream, overlay, call_id):
+    async def complete(self, *, messages, source, stream, overlay, call_id, reasoning_effort):
         self.calls.append((source.name, call_id))
+        self.efforts.append(reasoning_effort)
         action = self.script.pop(0)
         if isinstance(action, Exception):
             raise action
@@ -135,6 +143,37 @@ def _harness(
 _REQ = ChatRequest(messages=[{"role": "user", "content": "hi"}])
 
 
+class TestRequestTierReachesTransport:
+    """请求级档位必须一路穿过洋葱到达 transport(Task 5b)。
+
+    `ChatRequest` 上填了字段而中间件不搬运,是"看起来配了、实际没发出去"的静默
+    失效——正是 issue #20 里下游改用 extra_body 绕过治理的成因。
+    """
+
+    async def test_request_tier_reaches_transport(self):
+        mw, _, _, transport, *_ = _harness([_src("a")], [_ok()])
+        await mw(
+            ChatRequest(messages=[{"role": "user", "content": "hi"}], reasoning_effort=Effort.HIGH)
+        )
+        assert transport.efforts == [Effort.HIGH]
+
+    async def test_absent_tier_is_carried_as_none(self):
+        """不表态也要显式传下去: 漏传与"传了 None"在协议上必须区分不开才安全。"""
+        mw, _, _, transport, *_ = _harness([_src("a")], [_ok()])
+        await mw(_REQ)
+        assert transport.efforts == [None]
+
+    async def test_tier_is_carried_on_every_retry_attempt(self):
+        """换源重试时档位不得在第二次尝试上丢失。"""
+        mw, _, _, transport, *_ = _harness(
+            [_src("a"), _src("b")], [TransientError("boom", source_name="a"), _ok()]
+        )
+        await mw(
+            ChatRequest(messages=[{"role": "user", "content": "hi"}], reasoning_effort=Effort.LOW)
+        )
+        assert transport.efforts == [Effort.LOW, Effort.LOW]
+
+
 class TestSuccessPath:
     async def test_first_attempt_success_builds_response(self):
         mw, limiter, gate, transport, sleep, _ = _harness([_src("a")], [_ok("hello")])