Re-sample extensively for resulting projections
Only after #31 (closed) an #38 (closed) are done.
Samples found for each of the resulting projections are not correct for computation of marginal likelihood of data coming from model (Bayes factor ratio components), because they may underrepresent likelihood function domain when coming as projections from a larger model. Re-sampling is advised, possibly with larger sample size as for the filtering.
Allow to specify directly as an option to re-sample for best resulting projections. Best projections could be given by indication given by the volume of viable points used to find the projection (cf. the probabilities originally computed by Mikael; #31 (closed)). Think of threshold for how many of them to re-sample, e.g. all that satisfy a threshold for marginal likelihood indicator, at least 3 but not more than 10.