[PyTorch] Skip context parallelism tests if not enough GPUs

[PyTorch] Skip context parallelism tests if not enough GPUs #7076