[docs] Add working end-to-end example for a local model #40

It can be tricky to understand exactly how to use the different binaries to run the benchmark end-to-end.

I propose adding an example that the user can follow. Bonus points for making the guide platform agnostic so that the user is not dependent on being on Linux or needing a CUDA compatible GPU.