[Profiler][Trivial] Switch to nanoseconds for Result's internal representation
Pull Request resolved: https://github.com/pytorch/pytorch/pull/77697
Certain steps in building the call tree rely on sorting, so we want to retain as much precision as possible. `profiler_kineto.cpp` and KinetoEvent still use microseconds.
Differential Revision: [D36302563](https://our.internmc.facebook.com/intern/diff/D36302563/)
Approved by: https://github.com/aaronenyeshi
