fix --benchmark option

Created by: jieru-hu

Patch Description metaseq.launcher.opt_baselines's --benchmark option seems to be broken by this commit 1d2d1525

Testing steps

before the change

$ python -m metaseq.launcher.opt_baselines --model-size 8m --benchmark -t 1 -g 8 -n 1 -p test-8m1 --azure 
Traceback (most recent call last):
  File "/shared/home/jieru/miniconda3/envs/fairseq-20220503/lib/python3.9/runpy.py", line 197, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/shared/home/jieru/miniconda3/envs/fairseq-20220503/lib/python3.9/runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "/shared/home/jieru/fork/metaseq/metaseq/launcher/opt_baselines.py", line 342, in <module>
    cli_main()
  File "/shared/home/jieru/fork/metaseq/metaseq/launcher/opt_baselines.py", line 336, in cli_main
    sweep_main(
  File "/shared/home/jieru/fork/metaseq/metaseq/launcher/sweep.py", line 378, in main
    backend_main(get_grid, postprocess_hyperparams, args)
  File "/shared/home/jieru/fork/metaseq/metaseq/launcher/slurm.py", line 34, in main
    grid = get_grid(args)
  File "/shared/home/jieru/fork/metaseq/metaseq/launcher/opt_baselines.py", line 160, in get_grid
    hyperparam("--valid-subset", ",".join(f"valid/{ss}" for ss in valid_subsets)),
UnboundLocalError: local variable 'valid_subsets' referenced before assignment

after this commit

$ python -m metaseq.launcher.opt_baselines --model-size 8m --benchmark -t 1 -g 8 -n 1 -p test-8m1 --azure 
....

Launched job 84182
Launched 84182

Considerations before submitting:

• Was this discussed/approved via a Github issue?
• Did you read the contributor guideline?
• Did you make sure to update the docs?
• Did you write any new necessary tests? -->