fix gemini truncated reply (max_tokens 16->64) and suppress GeneratorExit on client disconnect

studio/backend/routes/inference.py

@@ -967,6 +967,8 @@ async def _proxy_to_external_provider(
             ):
                 yield f"{line}\n\n"
             yield "data: [DONE]\n\n"
+        except GeneratorExit:
+            pass
         except Exception as exc:
             logger.error("external_provider.stream_error", error = str(exc))
         finally:

studio/backend/tests/test_providers_api.py

@@ -512,7 +512,7 @@ class TestVisionInference:
                 ],
             }],
             "stream": True,
-            "max_tokens": 16,
+            "max_tokens": 64,
             "provider_type": provider_type,
             "external_model": model,
             "encrypted_api_key": encrypted,