Fix CUDA/DML package exception caused by ENABLE_CUDA_NHWC_OPS (#22851)

### Description
Now,  ENABLE_CUDA_NHWC_OPS is enabled by default.
It adds a new chance to create cuda provider while both cuda/dml are
enabled


### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->

onnxruntime/test/lora/lora_test.cc

@@ -201,6 +201,16 @@ TEST(LoraAdapterTest, Load) {
 
 #ifdef USE_CUDA
 TEST(LoraAdapterTest, VerifyDeviceCopy) {
+  // These checks for CUDA/DML combined Package, Be careful when you want to remove it!
+  if (DefaultCudaExecutionProvider() == nullptr) {
+    GTEST_SKIP() << "Skip This Test Due to this EP is null";
+  }
+#ifdef USE_DML
+  if (DefaultDmlExecutionProvider() != nullptr) {
+    GTEST_FAIL() << "It should not run with DML EP";
+  }
+#endif
+
   auto cpu_ep = DefaultCpuExecutionProvider();
   auto cpu_allocator = cpu_ep->CreatePreferredAllocators()[0];
   auto cuda_ep = DefaultCudaExecutionProvider();

onnxruntime/test/util/default_providers.cc

@@ -140,6 +140,12 @@ std::unique_ptr<IExecutionProvider> DefaultCudaExecutionProvider() {
 #ifdef ENABLE_CUDA_NHWC_OPS
 std::unique_ptr<IExecutionProvider> DefaultCudaNHWCExecutionProvider() {
 #if defined(USE_CUDA)
+#ifdef USE_DML
+  const std::string no_cuda_ep_test = Env::Default().GetEnvironmentVar("NO_CUDA_TEST");
+  if (no_cuda_ep_test == "1") {
+    return nullptr;
+  }
+#endif
   OrtCUDAProviderOptionsV2 provider_options{};
   provider_options.do_copy_in_default_stream = true;
   provider_options.use_tf32 = false;

tools/ci_build/github/azure-pipelines/templates/win-ci.yml

@@ -413,6 +413,7 @@ stages:
               workingDirectory: '$(Build.BinariesDirectory)'
             env:
               NO_CUDA_TEST: '1'
+              GTEST_FILTER: -*CudaNhwcTypedTest*
           - task: PythonScript@0
             displayName: 'test excludes DML'
             condition: and(succeeded(), eq('${{ parameters.runTests}}', true))