Simplify model benchmarking and deployment

[biswapanda]

Goals

• Make it easy to quickstart: One command to deploy and benchmark a model name, mode and config
• Make it repeatable: users can reproduce benchmark results and diagnose issues
• Make it Simple: Simplify the setup and decouple image and config from the benchmark code
• ProvideRecipies: Lay down golden path to deploy and benchmark few models with best known config based on tuning

[nnshah1]

Question: how does the relate to the work being done to generalize the k8s benchmarking - in some previous threads we shied away from using helm to deploy - so also want to understand