Feature: Allow switching of the payload if the metrics are far from the pareto front

[marenz2569]

We should be able to decrease the execution time of a payload when we detect, that this individual is far from the pareto front. This will decrease the accuracy of the model in these points, however we shouldn't need the accuracy around these points given that the model (metric = model(individual)) is monotonic and has no steep slopes. It should also mitigate thermal effects on the execution e.g., lower power consumption of subsequent individuals due to lower heat output of the current individual. This should improve the overall execution time of the optimization.

DoD

• Implement a way to switch between individuals faster if it is deemed impractical for further optimization (far away from the pareto front)
• Evaluate the change of accuracy of the build up model.
• Evaluate the change in total power consumptions vs. optimization time