LLAMA2 Simulated version

[Yash9060]

Fixes b/378909790
I have added a doc in the above issue which describes at how I calculated the correct size for each model.

There is lot of code duplication between multinode/simulated demo/benchmarks. I will raise a follow up PR to move all the code into common directory and re-use it everywhere.

I will also update documentation on how to run benchmark code in follow up PR.

Also removed certain piece of code from benchmark/simulated/benchmark.py which I accidentally checked in an earlier PR.

It's a good idea to open an issue first for discussion.

• Tests pass

[awonak]

Nice work, thanks for the additional documentation and links to articles for context!