Fix FakeTensor device creation for MPS

[malfet]

Stack from ghstack (oldest at bottom):

• [MPSInductor] Implement pow() #144827
• [MPSInductor] Support abs in MetalPrintExpr #144826
• -> Fix FakeTensor device creation for MPS #144796
• [BE] Extend test_remove_no_ops #144795

By promoting torch.device("mps") to torch.device("mps:0"), but skipping is_initialized check, as MPS does not really support multi-GPU right now

This fixes GPUTests.test_remove_no_ops_mps

cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire @chauhang @aakhundov @BoyuanFeng

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/144796

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure, 2 Pending, 1 Unrelated Failure

As of commit c799cef with merge base 95b41d2 ():

NEW FAILURE - The following job has failed:

• pull / linux-jammy-py3.10-clang15-asan / test (default, 5, 6, linux.4xlarge) (gh)
  Process completed with exit code 137.

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• pull / linux-jammy-py3-clang12-executorch / test (executorch, 1, 1, linux.2xlarge) (gh) (trunk failure)
  backends/xnnpack/test/ops/test_conv1d.py::TestConv1d::test_qs8_conv1d_batchnorm_seq

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[ezyang]

what's going on here?

[malfet]

Is it documented anywhere, that torch.device.is_initialized() should be available? I can implement that function, but I'm unsure what it actually means. That memory were allocated on GPU?

[ezyang]

I mean, we have torch.cuda.is_initialized which means the CUDA context has been initialized. If your accelerator has no meaningful concept of initialization you can just always return True.