[need] iterative refinement in the MA27/MA57 wrapper classes of PIPS-NLP

[cnpetra]

strongly recommended. for best performance, this should be done for multiple rhs along the lines of @BarrySmith 's comment from PR #58