v00 : a bit smarter than thesis
     reviewing of implementation of Okulov and Sorensen/Okulov (debugging paper mistakes ), two papers from 2007

v01: smarter Goldstein function computation (no loop on blades)
     Used matlab coder with upper bound of 3000 points for fast Goldstein computation   
    Seem to fail though for really small h (if I recall right)

v01_nomex not really a version but avoid using the mex

