[PyTorch] Skip context parallelism tests if not enough GPUs (#1508)

[PyTorch] Skip context parallelism tests if not enough GPUs (#1508) #93