Stochastic predictions for binary classification

[MorrisNein]

Some models appear to behave differently in different environments.

For example, a pipeline consisting of multiple logistic regressions gives different outputs for the same data:

https://github.com/aimclub/FEDOT/actions/runs/7086728823/job/19285547590?pr=1214

This does not allow testing binary classification values (in the test test_metrics). The source of the stochastic behaviour should be eliminated, at least for the test.