Benchmark aotautograd+nvfuser (#932)
Summary:
Add support to run nvfuser and nvfuser+aotautograd (aotautograd_speedup_strategy in torchdynamo) backends in the workflow config.
Pull Request resolved: https://github.com/pytorch/benchmark/pull/932
Test Plan:
```
$ python run_sweep.py -d cuda -t train --torchdynamo aot_autograd_speedup_strategy
```
Reviewed By: jansel
Differential Revision: D36852083
Pulled By: xuzhao9
fbshipit-source-id: 46a63ecf4604218abe867987377e27d053e80547
