[Feature] Support Pipeline Parallism in torchrun SPMD offline inference for V1 (#17827)

Signed-off-by: Lucia Fang <fanglu@fb.com>

vllm/config.py

@@ -1695,7 +1695,6 @@ class ParallelConfig:
     """Port of the data parallel master."""
     enable_expert_parallel: bool = False
     """Use expert parallelism instead of tensor parallelism for MoE layers."""
-
     max_parallel_loading_workers: Optional[int] = None
     """Maximum number of parallel loading workers when loading model
     sequentially in multiple batches. To avoid RAM OOM when using tensor