[op_collection] Force --isolate mode when running multiple ops

[xuzhao9]

Correctly run op_collection with --isolate.
Force using --isolate mode when running more than one ops.

Fixes #107

Test plan:

$ python run.py --op-collection liger --mode fwd_bwd --precision fp32 --metrics latency,speedup --dump-csv --isolate
       (B, T, V)    torch_kl_div-latency    liger_kl_div-speedup    liger_kl_div-latency    inductor_kl_div-speedup    inductor_kl_div-latency
----------------  ----------------------  ----------------------  ----------------------  -------------------------  -------------------------
  (8, 512, 4096)                0.494784                 1.40525                0.352096                    1.92266                   0.257344
  (8, 512, 8192)                0.931776                 1.49861                0.62176                     1.85099                   0.503392
 (8, 512, 16384)                1.81686                  1.54794                1.17373                     1.82287                   0.996704
 (8, 512, 32768)                3.66314                  1.55627                2.35379                     1.7796                    2.0584
 (8, 512, 65536)                7.2864                   1.57052                4.63949                     1.76966                   4.11741
(8, 512, 131072)               14.5042                   1.57413                9.21411                     1.76836                   8.20205

[facebook-github-bot]

@xuzhao9 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[FindHao]

LGTM! Thanks for fixing it!

[facebook-github-bot]

@xuzhao9 merged this pull request in 1b51626.