Add userbenchmark for Torch-TRT Backend (#1811)
Summary:
- Add handling for benchmarking multiple models in a single run using `ModelTask` instances
- Add error catching for testing all models and producing metrics
- Add CI yaml file to specify run configuration
Pull Request resolved: https://github.com/pytorch/benchmark/pull/1811
Reviewed By: FindHao
Differential Revision: D48229179
Pulled By: xuzhao9
fbshipit-source-id: 3d6fc5c11fa7ff081f25ca0d185ca828fc3085db
