Consider other options for performance benchmarks

[gshep]

File Location(s)

No response

Proposal

Caused by #900

The following options can be considered as a further evolvement of the task:

1. choose some algorithm as a base measurement of hardware state. Run it on the main branch and the source PR branch then use the results to adjust benchmark values;
2. use box-plots for both the main branch and the feature/working branch. Since collecting statistics is a time consuming operation we could create some label and perform the operation only for labeled PRs;
3. add new standalone runner. All performance benchmarks analysis should be done on the runner only.