Add "cluster" option for slurm cluster settings (#1362)
Summary:
Pull Request resolved: https://github.com/pytorch/benchmark/pull/1362
This allows you to run with `--cluster local`, to utilize the GPUs on the current allocation (so you don't have to launch a new slurm job or have a pytorch env that's accessible across multiple machines)
Test Plan: Imported from OSS
Reviewed By: wconstab
Differential Revision: D42437977
Pulled By: davidberard98
fbshipit-source-id: 4cddde819c638f18d421b8f075e76e35d40b648f
