erfcx_xpu and ndtri_xpu not implemented for 'BFloat16'

[huaiyuzh]

erfcx_xpu and ndtri_xpu cause an IPEX UT fail. In IPEX2.6, we override this Ops with IPEX implementation to make this UT pass.
ipex/tests/gpu/example/test_special_ops.py::TestTorchMethod::test_erfcx - RuntimeError: "erfcx_xpu" not implemented for 'BFloat16'
ipex/tests/gpu/example/test_special_ops.py::TestTorchMethod::test_ndtri_entr - RuntimeError: "ndtri_xpu" not implemented for 'BFloat16' True

[chunhuanMeng]

This is because ipex adds support for BFloat16 data type for these two ops, but torch-xpu-ops does not have this support, and stock pytorch also does not have this support. Therefore, if you really need it, you can raise a PR in pytorch, and then torch-xpu-ops will make corresponding changes. We ensure that our code is consistent with the design of stock pytorch.
ref link:
torch-xpu-ops：

torch-xpu-ops/src/ATen/native/xpu/sycl/UnarySpecialOpsKernels.cpp

Line 304 in 306a0ff

AT_DISPATCH_FLOATING_TYPES(iter.common_dtype(), "erfcx_xpu", [&]() {

torch-xpu-ops/src/ATen/native/xpu/sycl/UnarySpecialOpsKernels.cpp

Line 255 in 306a0ff

AT_DISPATCH_FLOATING_TYPES(iter.common_dtype(), "ndtri_xpu", [&]() {

cuda:
https://github.com/pytorch/pytorch/blob/bb7e8fbd668c7c8931436b4a935b26911cbe0daf/aten/src/ATen/native/cuda/UnarySpecialOpsKernel.cu#L310;
https://github.com/pytorch/pytorch/blob/bb7e8fbd668c7c8931436b4a935b26911cbe0daf/aten/src/ATen/native/cuda/UnarySpecialOpsKernel.cu#L230
ipex:
https://github.com/intel/intel-extension-for-pytorch/blob/release/xpu/2.6.10/csrc/gpu/aten/operators/SpecialOps.cpp#L52;
https://github.com/intel/intel-extension-for-pytorch/blob/release/xpu/2.6.10/csrc/gpu/aten/operators/SpecialOps.cpp#L65