[FEA]: Allow multiple benchmarks in a single tuning

[gevtushenko]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this request and that I agree to the Code of Conduct

Area

CUB

Is your feature request related to a problem? Please describe.

Currently, the tuning infrastructure only supports a single benchmark per file. This makes it difficult to organize a search process on disjoint sets of workloads produced by Cartesian product of runtime axes.

Describe the solution you'd like

Tasks

Preview Give feedback

1. Allow multiple benchmarks in a single tuning
2. Extend batched memcpy benchmark with new workload covering small number of very large segments

Describe alternatives you've considered

We could use skip feature along with including all possible options in a single Cartesian product, but this creates a lot of noise.

Additional context

No response