fix(BIZ-26): 限流范围收窄到 NVIDIA 网关

- 新增网关识别逻辑：只识别 nvidia / nvidiavx18088980513 为限流目标
- volcengine-plan、siliconflow、deepseek 等非 NVIDIA 网关默认不进入令牌桶
- RequestScheduler 增加 gateway/model 参数与 _should_rate_limit 判断
- 未知网关默认不限流，避免误伤其他通道
- 补充网关范围测试与使用文档说明

Co-authored-by: multica-agent <github@multica.ai>

scripts/test_rate_limiter.py

@@ -28,6 +28,7 @@ from rate_limiter import (
     Priority,
     retry_with_backoff,
     CoordinatedPoller,
+    is_nvidia_gateway,
 )
 
 
@@ -240,9 +241,10 @@ def test_rate_limit_stress():
     for i in range(50):
         priority = Priority.NORMAL if i % 10 != 0 else Priority.URGENT
         scheduler.submit(
-            payload={"index": i},
+            payload={"index": i, "provider": "nvidia"},
             priority=priority,
-            callback=callback
+            callback=callback,
+            gateway="nvidia"
         )
     
     print("提交完成，等待处理...")
@@ -266,6 +268,29 @@ def test_rate_limit_stress():
     print("\n✅ 压力测试完成\n")
 
 
+def test_gateway_scope():
+    """测试限流范围：只对 NVIDIA 网关生效"""
+    print("=" * 60)
+    print("测试 7: 网关范围识别（只限 NVIDIA）")
+    print("=" * 60)
+
+    assert is_nvidia_gateway("nvidia") is True
+    assert is_nvidia_gateway("nvidiavx18088980513/deepseek-ai/deepseek-v4-pro") is True
+    assert is_nvidia_gateway("volcengine-plan/ark-code-latest") is False
+    assert is_nvidia_gateway("siliconflow/Qwen/Qwen3") is False
+    assert is_nvidia_gateway("deepseek/deepseek-chat") is False
+    assert is_nvidia_gateway(None) is False
+
+    scheduler = RequestScheduler(rate=1/60, capacity=1, enable_cache=True)
+    # 先耗尽 NVIDIA 桶
+    scheduler.submit(payload={"provider": "nvidia", "i": 1}, priority=Priority.NORMAL, callback=lambda x: x, gateway="nvidia")
+    # 非 NVIDIA 请求应直接执行，不受桶状态影响
+    non_nv = {"provider": "volcengine-plan", "i": 2}
+    assert scheduler._should_rate_limit(type("R", (), {"gateway": "volcengine-plan", "model": None, "payload": non_nv})()) is False
+
+    print("✅ 网关范围识别测试完成：volcengine-plan/siliconflow/DeepSeek 不限流，NVIDIA 限流\n")
+
+
 def main():
     """运行所有测试"""
     print("\n")
@@ -284,6 +309,7 @@ def main():
         test_retry_decorator()
         test_coordinated_poller()
         test_rate_limit_stress()
+        test_gateway_scope()
         
         print("\n")
         print("╔" + "=" * 58 + "╗")