Add performance profiling tool to enable programmatic/interactive/automatic PT-XLA profile capturing (#3498)
* Add performance profiling tool to enable programmatic/interactive/automatic PT-XLA profile capturing
* Update README to document capture_profile.py
* Add module docstring to capture_profile.py
* Add instructions on how to run tensorboard in module docstring for capture_profile.py
* Reformat capture_profile.py using yapf