fix: address M2.5 verifier findings before merge

AIMD ceiling now respects per-source max_concurrency and the pacer is
assembled explicitly in the client; MIN_CALLS parses as strict int;
acceptance doc corrects source-5 attempt count to 549; design and
migration notes aligned with implemented 429/stall/suppression
semantics and AIMD constants documented.

tests/unit/test_health_selector.py

@@ -119,3 +119,48 @@ class TestAdaptivePacer:
         assert pacer.admit("s1") is True
         pacer.leave("s1")  # 多余 leave 不下穿 0
         assert pacer._inflight.get("s1", 0) == 0
+
+
+class TestPacerAssembly:
+    """独立核验 I1/M5: ceiling 尊重源级并发;取消路径在途归零。"""
+
+    def test_ceiling_respects_large_source_concurrency(self):
+        from polygateway.client import GatewayClient
+        from polygateway.config import GatewaySettings
+
+        env = {
+            "LLM__QWEN__1__BASE_URL": "https://gw.example/v1",
+            "LLM__QWEN__1__API_KEY": "sk-a",
+            "LLM__QWEN__1__MODEL": "m",
+            "LLM__QWEN__1__TIMEOUT_S": "60",
+            "LLM__QWEN__1__MAX_CONCURRENCY": "128",
+            "LLM_MAX_RETRIES": "3",
+            "LLM_RETRY_BASE_DELAY": "2.0",
+            "LLM_RETRY_MAX_DELAY": "30.0",
+            "LLM_CIRCUIT_BREAKER_THRESHOLD": "5",
+            "LLM_CIRCUIT_BREAKER_COOLDOWN": "60",
+            "PGW_CACHE_BACKEND": "none",
+            "PGW_TELEMETRY_BACKEND": "none",
+        }
+        client = GatewayClient.from_settings(GatewaySettings.from_env("LLM", env=env))
+        assert client._terminal._pacer._ceiling == pytest.approx(128.0)
+
+    def test_min_calls_rejects_float_value(self):
+        from polygateway.config import GatewaySettings
+
+        env = {
+            "LLM__QWEN__1__BASE_URL": "https://gw.example/v1",
+            "LLM__QWEN__1__API_KEY": "sk-a",
+            "LLM__QWEN__1__MODEL": "m",
+            "LLM__QWEN__1__TIMEOUT_S": "60",
+            "LLM_MAX_RETRIES": "3",
+            "LLM_RETRY_BASE_DELAY": "2.0",
+            "LLM_RETRY_MAX_DELAY": "30.0",
+            "LLM_CIRCUIT_BREAKER_THRESHOLD": "5",
+            "LLM_CIRCUIT_BREAKER_COOLDOWN": "60",
+            "LLM__BREAKER__MIN_CALLS": "10.5",
+            "PGW_CACHE_BACKEND": "none",
+            "PGW_TELEMETRY_BACKEND": "none",
+        }
+        with pytest.raises(ValueError, match="MIN_CALLS"):
+            GatewaySettings.from_env("LLM", env=env)