fix: map Zhipu GLM non-standard finish_reason values

Zhipu GLM returns non-standard finish_reason values during streaming
when inference fails mid-request, causing Pydantic validation crash:
- "network_error" (inference interrupted) → map to "stop"
- "sensitive" (content policy violation) → map to "content_filter"

Fixes #23386

litellm/litellm_core_utils/core_helpers.py

@@ -89,6 +89,9 @@ _FINISH_REASON_MAP: dict[str, OpenAIChatCompletionFinishReason] = {
     "IMAGE_PROHIBITED_CONTENT": "content_filter",
     "TOO_MANY_TOOL_CALLS": "stop",
     "MALFORMED_RESPONSE": "stop",
+    # Zhipu GLM
+    "network_error": "stop",
+    "sensitive": "content_filter",
     # Bedrock
     "guardrail_intervened": "content_filter",
     # OpenAI passthrough

tests/test_litellm/litellm_core_utils/test_core_helpers.py

@@ -126,6 +126,14 @@ class TestMapFinishReasonBedrock:
         assert map_finish_reason("guardrail_intervened") == "content_filter"
 
 
+class TestMapFinishReasonZhipu:
+    def test_network_error(self):
+        assert map_finish_reason("network_error") == "stop"
+
+    def test_sensitive(self):
+        assert map_finish_reason("sensitive") == "content_filter"
+
+
 class TestMapFinishReasonOpenAIPassthrough:
     @pytest.mark.parametrize(
         "reason", ["stop", "length", "tool_calls", "function_call", "content_filter"]