enhance: dimension of hot scheduler

[lhy1024]

Feature Request

Describe your feature request related problem

After some tests, since each query consumes different resources, if we just try to make the number of queries consistent, there is a possibility that the number of queries is the same between each store, but the cpu is different.

Also, this may have something to do with the distribution of tables.

Describe the feature you'd like

Perhaps we need to give feedback on scheduling based on cpu, or consider the dimension of the table, or we give them different weights

Describe alternatives you've considered

Teachability, Documentation, Adoption, Migration Strategy

[lhy1024]

Can we consider adding the factor of cpu when dealing with the conflict of the two dimensions? @HunDunDM

[lhy1024]

Although qps and byte are both uniform, but their cpu is different

We can collect unified pool cpu by query from tikv, but cannot collect grpc cpu by region and query now. And we still need to collect cpu about write.