[xpu][fix]Fall back deterministic index_copy to index_put on XPU (pytorch#167830)

A minor update has been made to the deterministic behavior checks in the `index_copy_out` implementation. This change ensures that deterministic  `index_copy` is dispatched to `index_put` not only for CUDA tensors but also for XPU tensors.

Pull Request resolved: pytorch#167830
Approved by: https://github.com/guangyey, https://github.com/ezyang

Author: chunhuanMeng

aten/src/ATen/native/TensorAdvancedIndexing.cpp

@@ -1087,7 +1087,8 @@ TORCH_IMPL_FUNC(index_copy_out)
     result.copy_(self);
 
   // See Note [Enabling Deterministic Operations]
-  if (result.is_cuda() && globalContext().deterministicAlgorithms()) {
+  if ((result.is_cuda() || result.is_xpu()) &&
+      globalContext().deterministicAlgorithms()) {
     torch::List<std::optional<Tensor>> indices;
     indices.resize(dim + 1);
     indices.set(dim, index);