[webgpu] Expand Unsqueeze version to 23 (microsoft#25858)

### Description
The phi4 mini in Edge is using ai.onnx v21. Without this change, it
results a `MemcpyToHost` inserted and slows the generation speed.

Author: qjia7

onnxruntime/core/providers/webgpu/tensor/unsqueeze.cc

@@ -11,7 +11,31 @@ namespace webgpu {
 ONNX_OPERATOR_KERNEL_EX(
     Unsqueeze,
     kOnnxDomain,
-    13,
+    23,
+    kWebGpuExecutionProvider,
+    (*KernelDefBuilder::Create())
+        .TypeConstraint("T", WebGpuSupportedNumberTypes())
+        .TypeConstraint("axes", DataTypeImpl::GetTensorType<int64_t>())
+        .Alias(0, 0)
+        .InputMemoryType(OrtMemTypeCPU, 1),
+    Unsqueeze);
+
+ONNX_OPERATOR_VERSIONED_KERNEL_EX(
+    Unsqueeze,
+    kOnnxDomain,
+    21, 22,
+    kWebGpuExecutionProvider,
+    (*KernelDefBuilder::Create())
+        .TypeConstraint("T", WebGpuSupportedNumberTypes())
+        .TypeConstraint("axes", DataTypeImpl::GetTensorType<int64_t>())
+        .Alias(0, 0)
+        .InputMemoryType(OrtMemTypeCPU, 1),
+    Unsqueeze);
+
+ONNX_OPERATOR_VERSIONED_KERNEL_EX(
+    Unsqueeze,
+    kOnnxDomain,
+    13, 20,
     kWebGpuExecutionProvider,
     (*KernelDefBuilder::Create())
         .TypeConstraint("T", WebGpuSupportedNumberTypes())

onnxruntime/core/providers/webgpu/webgpu_execution_provider.cc

@@ -249,7 +249,9 @@ class ONNX_OPERATOR_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 13,
 
 class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 1, 10, Unsqueeze);
 class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 11, 12, Unsqueeze);
-class ONNX_OPERATOR_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 13, Unsqueeze);
+class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 13, 20, Unsqueeze);
+class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 21, 22, Unsqueeze);
+class ONNX_OPERATOR_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 23, Unsqueeze);
 
 class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 9, 15, Where);
 class ONNX_OPERATOR_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 16, Where);
@@ -548,7 +550,9 @@ std::unique_ptr<KernelRegistry> RegisterKernels() {
 
       BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 1, 10, Unsqueeze)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 11, 12, Unsqueeze)>,
-      BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 13, Unsqueeze)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 13, 20, Unsqueeze)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 21, 22, Unsqueeze)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 23, Unsqueeze)>,
 
       BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 1, 10, ReduceMin)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kWebGpuExecutionProvider, kOnnxDomain, 11, 11, ReduceMin)>,