Add neuroSMT feature in calculation of reward

[alexeyfridman]

SAT-1641

We need to add new feature for executionState that estimate time of SMT request, then do experiment: train models and compare coverage on test project (guava-26.0).

[amandelpie]

This is an unfinished part of the Maxim Petukhov and other research, unfortunately there is no open addition information to add here