Solve differences in `TrackBestCreator` Ghost Probability counters

The hive-multi-thread Brunel test with the current master shows a difference in the Ghost Probability counters of the TrackBestCreator algorithm:

    (TrackBestTrackC ref) Downstream.ghostProbability | 3119 | 1823.269 | 0.58457
    (TrackBestTrackC new) Downstream.ghostProbability | 3119 | 1822.232 | 0.58424
    (TrackBestTrackC ref) VeloBackward.ghostProbability | 7448 | 459.089 | 0.061639
    (TrackBestTrackC new) VeloBackward.ghostProbability | 7448 | 460.1623 | 0.061783
    (TrackBestTrackC ref) Long.ghostProbability | 6753 | 1650.696 | 0.24444
    (TrackBestTrackC new) Long.ghostProbability | 6753 | 1655.471 | 0.24515
    (TrackBestTrackC ref) Velo.ghostProbability | 5889 | 548.2368 | 0.093095
    (TrackBestTrackC new) Velo.ghostProbability | 5889 | 549.1184 | 0.093245