Introduce is_big_gpu condition for test_max_autotune (#111467)
Fixes https://github.com/pytorch/pytorch/issues/111527
Other test files that rely on `max_autotune` mode being enabled already conditionalise the UT suite on this condition (e.g. test_select_algorithm). Proposing to add this condition for test_max_autotune.
Currently we are observing failures on these UTs on the ROCm runners but using MI200+ these tests will pass again (context: https://github.com/pytorch/pytorch/pull/111381#issuecomment-1768048732)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/111467
Approved by: https://github.com/shunting314