Fix skip for test_set_nccl_pg_timeout (#120130)
Test is failing on our internal CI with below error
```RuntimeError: CUDA error: invalid device ordinal
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
```
Purpose of this test is for nccl so it doesnt make sense to run in 1 GPU setting either.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120130
Approved by: https://github.com/wconstab, https://github.com/eqy
