Fix the skip condition for test_c10d tests (#119938)
Seeing the error for c10d tests when running on 1GPU. Adding the skip when there is insufficient GPU.
```
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.
```
referring to https://github.com/pytorch/pytorch/pull/84980
Pull Request resolved: https://github.com/pytorch/pytorch/pull/119938
Approved by: https://github.com/eqy, https://github.com/fegin
