[CK2/CM project] implement modular ML benchmarking based on loadgen and CK2(CM)

We want to develop a CK2(CM) script to enable modular ML benchmarking across diverse ML models, data sets, engines, libraries and platforms (available as CM scripts too) based on C++ loadgen as a part of Design Space Exploration activity within MLCommons. This ticket requires further discussion during our regular conf-calls ...

This is a part of the Q3 project: https://github.com/mlcommons/ck/issues/261

Moved to https://github.com/mlcommons/ck/issues/265