[Results] Improve stats for RTA PerfIndex metric
Created by: derkling
The current PerfIndex metric reports the average with respect to all the tasks composing the workload: https://github.com/ARM-software/lisa/blob/master/libs/utils/results.py#L253
This could introduce big variances in case different tasks have quite different averages. A better approach could be that to compute the average for each task independently.
A single metrics for the PerfIndex of an experiment can than be defined by ordering the per-task averages and comparing with respect to the same ordered list of the base case. This should be enough to build a single index which better reports the performance index variation with respect to the base experiment.