Towards Workload-Agnostic Solutions to Make Smarter Peak Power, Performance and Energy Trade-Offs when Scheduling Workloads on Heterogeneous Clusters