fix zero size input for where op on webgpu (#26576)

fixes an issue with edge-tam-encoder reported by hf

Author: guschmue

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

@@ -126,6 +126,10 @@ Status Where::ComputeInternal(ComputeContext& context) const {
   TensorShape output_shape;
   ORT_RETURN_IF_ERROR(ComputeOutputShape(cond_shape, x_shape, y_shape, output_shape));
   auto* output_tensor = context.Output(0, output_shape);
+  if (output_tensor->Shape().Size() == 0) {
+    return Status::OK();
+  }
+
   constexpr int component = 4;
   uint32_t vec_size = onnxruntime::narrow<uint32_t>((output_shape.Size() + 3) / component);
   const auto is_broadcast = !(x_shape == y_shape &&