No need to run non-CUDA jobs in memory leak check mode (#102188)
Memory leak check mode is only meant for runner with GPU such as CUDA and ROCm https://github.com/pytorch/pytorch/blob/main/torch/testing/_internal/common_utils.py#L1093. So it's a waste of time and resource to run them for CPU-only jobs
### Testing
CUDA jobs have both `mem_leak_check` and `rerun_disabled_tests`:
* https://github.com/pytorch/pytorch/actions/runs/5072143038/jobs/9109448858#step:9:131
* https://github.com/pytorch/pytorch/actions/runs/5072143038/jobs/9109449417#step:9:123
* https://github.com/pytorch/pytorch/actions/runs/5072143038/jobs/9109452338#step:9:111
Same goes for Bazel CUDA job:
* https://github.com/pytorch/pytorch/actions/runs/5072143038/jobs/9109451535#step:3:132
And ROCM job:
* https://github.com/pytorch/pytorch/actions/runs/5072143038/jobs/9109451353#step:9:117
Non CUDA or ROCM jobs have only `rerun_disabled_tests` mode, for example:
* https://github.com/pytorch/pytorch/actions/runs/5072143038/jobs/9109449894#step:9:127
* ASAN https://github.com/pytorch/pytorch/actions/runs/5072143038/jobs/9109449157#step:9:126
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102188
Approved by: https://github.com/clee2000
