1. Replace SUMSL with LBFGS everywhere (currently only minim_ecart)
2. Parallelize the transformation of Cartesian derivatives to backbone-angle
   derivatives (cart2int.F)
3. Parallelize usampl
4. HREMD 
5. Fourbody interactions - eliminate pp-contact etc. storage (currently 4body 
   switched off by default, -DFOURBODY flag switches on).
6. Fix LBFGS - the linesearch (search) subroutine does not handle difficult
   cases when changing variables can result in nans.