support for 'ray job submit --working-dir . -- python start_xxx_pipeline.py ' start

目前多机训练需要为每台机器配置环境变量，并运行python start_xxx_pipeline.py启动训练。
希望新增 ray job submit --working-dir . -- python start_xxx_pipeline.py 启动方式，这样就可以直接在已有的 ray cluster 上运行任务了。

Start by tracing how the existing multi-machine training command is launched and where the training entry point is selected. Verify the requested `ray job submit --working-dir . -- python start_xxx_pipeline.py` invocation on an existing Ray cluster, with completion meaning the job starts without per-machine environment configuration.

Written by the indexing model from the issue text.