Enable two XPU dtype paths that were causing quantization tests to fail.

[BBBela]

Fixes: #4445

Changes

1. Add Float4_e2m1fn_x2 support to XPU eq/ne compare dispatch.
2. Add Bits* support to XPU copy.

Motivation

Without this change, two tests from upstream are failing:

python test/quantization/core/experimental/test_floatx.py TestFloat4DtypeXPU.test_float4_e2m1fn_x2_xpu
python test/quantization/core/experimental/test_bits.py TestBitsXPU.test_types_xpu

this change fixes those two cases and allows to enable xpu on those two test files.

[BBBela]

Thank you @tszulist-hbn and @astachowiczhabana for spotting the lack of TORCH_CHECK! I added this check like CUDA does.

[BBBela]

For the record - after adding allow_xpu=True to the upstream test files, the tests are passing with changes from this PR:

~/pytorch/test$ python -m pytest -v quantization/core/experimental/test_bits.py -k test_types_xpu
=============================================================================== test session starts ===============================================================================

quantization/core/experimental/test_bits.py::TestBitsXPU::test_types_xpu PASSED [0.1433s]                                                                                    [100%]

~/pytorch/test$ python -m pytest -v quantization/core/experimental/test_floatx.py -k test_float4_e2m1fn_x2_xpu
=============================================================================== test session starts ================================================================================

quantization/core/experimental/test_floatx.py::TestFloat4DtypeXPU::test_float4_e2m1fn_x2_xpu PASSED [0.4205s]                                                                [100%]

[BBBela]

@torchxpubot merge

[torchxpubot]

PR has been successfully merged by @BBBela.