mpi_new_best_fit_hook misses actual best fits
When using mpi_new_best_fit_hook to, e.g., save the found best fit parameters that are found during mpi_fit_pars most of the time it is not the actual best fit that occurs during confidence search.
It seems that only the master process is executing mpi_new_best_fit_hook but some of the best fits found by the child processes are not reported to master or incorrectly.