Enable FLOPS Computation for Experimental Kineto Profiler (#51503)
Summary:
Add the FLOPS metric computation to the experimental Kineto profiler.
This includes saving necessary extra arguments and compute flops in the C++ code,
and extract the FLOPS value from the Python frontend.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/51503
Test Plan:
Build PyTorch with USE_KINETO option, then run the unit test:
```python
python test/test_profiler.py -k test_flops
```
Reviewed By: ilia-cher
Differential Revision: D26202711
Pulled By: xuzhao9
fbshipit-source-id: 7dab7c513f454355a220b72859edb3ccbddcb3ff
