Using the latest device-agnostic api in Context (#934)

There are some code refactorings for Context in PyTorch repo, refer to
[this](pytorch/pytorch#136519) to get more info,
we want to merge this pr in torch-xpu-ops first and then update the
pin-commit in PyTorch to make the ci workflow green in PyTorch.

src/ATen/native/xpu/sycl/ResizeKernel.cpp

@@ -25,7 +25,6 @@ void resize_bytes_xpu(StorageImpl* storage, size_t size_bytes) {
   c10::xpu::XPUGuard guard(device.index());
   at::DataPtr data = allocator->allocate(size_bytes);
   if (storage->data_ptr()) {
-    at::globalContext().lazyInitXPU();
     auto q = at::xpu::getCurrentSYCLQueue();
 
     q.memcpy(