[windows CI]open inference_ut in windows-inference pipeline (#43446)

* open inference_ut;test=windows_ci_inference

* inference_ut need onnx;test=windows_ci_inference

* disable trt_split_converter_test; use higher parallel level

* too high parallel will cause ut timeout

paddle/scripts/paddle_build.bat

@@ -680,7 +680,12 @@ pip install requests
 
 set PATH=%THIRD_PARTY_PATH:/=\%\install\openblas\lib;%THIRD_PARTY_PATH:/=\%\install\openblas\bin;^
 %THIRD_PARTY_PATH:/=\%\install\zlib\bin;%THIRD_PARTY_PATH:/=\%\install\mklml\lib;^
-%THIRD_PARTY_PATH:/=\%\install\mkldnn\bin;%THIRD_PARTY_PATH:/=\%\install\warpctc\bin;%PATH%
+%THIRD_PARTY_PATH:/=\%\install\mkldnn\bin;%THIRD_PARTY_PATH:/=\%\install\warpctc\bin;^
+%THIRD_PARTY_PATH:/=\%\install\onnxruntime\lib;%THIRD_PARTY_PATH:/=\%\install\paddle2onnx\lib;^
+%work_dir%\%BUILD_DIR%\paddle\fluid\inference;%PATH%
+
+REM TODO: make ut find .dll in install\onnxruntime\lib
+xcopy %THIRD_PARTY_PATH:/=\%\install\onnxruntime\lib\onnxruntime.dll %work_dir%\%BUILD_DIR%\paddle\fluid\inference\tests\api\ /Y
 
 if "%WITH_GPU%"=="ON" (
     call:parallel_test_base_gpu

tools/windows/run_unittests.sh

@@ -85,6 +85,7 @@ disable_win_inference_api_test="^trt_quant_int8_yolov3_r50_test$|\
 ^lite_resnet50_test$|\
 ^test_trt_dynamic_shape_transformer_prune$|\
 ^lite_mul_model_test$|\
+^trt_split_converter_test$|\
 ^paddle_infer_api_copy_tensor_tester$"
 
 
@@ -191,10 +192,6 @@ if [ -f "$PADDLE_ROOT/added_ut" ];then
         echo "========================================"
         exit 8;
     fi
-    if nvcc --version | grep 11.2; then
-        echo "Only test added_ut temporarily when running in CI-Windows-inference of CUDA 11.2."
-        exit 0;
-    fi
 fi
 set -e