Create a benchmark/test to measure convergence of p_non_comp learning

[meneguzzi]

Currently, @scranefield created a scenario to test the learning of p_non_comp in
Norm Detector benchmark

We want to create a benchmark to measure the convergence of learning p_non_comp learning, so perhaps we should move this to a completely new method in the benchmark.

The implementation of the learning is
update_p_non_comp

UpdateOddsRatioVsNoNorm
calls update_p_non_comp leading to
update_p_non_comp