Merge PR #98/#100 (feat #57 InferenceBackend 抽象契约 + #58 GPU Triton 后端，桥接 #44/#45 本地与云端实现)

core/llm-gateway/__init__.py

@@ -49,13 +49,21 @@ from .hallucination import (
     HallucinationGuard,
 )
 from .gateway import (
-    CloudBackend,
     GatewayResult,
-    InferenceBackend,
     LLMGateway,
-    LocalBackend,
 )
-from .backends import CloudApiBackend, Local70BBackend
+from .backends import (
+    BackendCapabilities,
+    BackendHealth,
+    CloudApiBackend,
+    CloudBackend,
+    InferResult,
+    InferenceBackend,
+    Local70BBackend,
+    LocalBackend,
+    build_backend,
+    default_registry,
+)
 
 __all__ = [
     # dlp
@@ -67,9 +75,10 @@ __all__ = [
     # hallucination
     "GuardVerdict", "HallucinationGuard",
     # gateway
-    "InferenceBackend", "LocalBackend", "CloudBackend",
-    # backends
-    "Local70BBackend",
-    "CloudApiBackend",
     "GatewayResult", "LLMGateway",
+    # 推理后端（#57 抽象契约 + #44 本地 / #45 云端 / #58 GPU）
+    "InferenceBackend", "BackendCapabilities", "BackendHealth", "InferResult",
+    "LocalBackend", "CloudBackend",
+    "Local70BBackend", "CloudApiBackend",
+    "default_registry", "build_backend",
 ]