Resurrect ao benchmark on AWS A100 runner (#2561)
Summary:
I'm bringing back some example models first, one for each set (TIMM, HF, TorchBench), to have some data to unblock our TorchAO ClickHouse migration. More can be added later if we decide to keep this workflow.
### Testing
https://github.com/pytorch/benchmark/actions/runs/12388956432/job/34581035274
The results are now available on `oss_ci_benchmark_v3` table `select * from oss_ci_benchmark_v3 where workflow_id = 12388956432`
Pull Request resolved: https://github.com/pytorch/benchmark/pull/2561
Reviewed By: xuzhao9
Differential Revision: D67383198
Pulled By: huydhn
fbshipit-source-id: cbe3e4c6517c656981c7e00d8564451a4d3654d0
