New RMSD implementation

RMSD re-implemented from scracth for the case where align==displace.
The new routine is much faster, but has some problem of numerical stability
(rt32 is broken). The code is choosing automatically between new version
(when align==displace) and old version (when align!=displace) so as
to allow the latter for e.g. docking. In RMSD.cpp, you can edit these lines:
// set to true one of these to enforce old routine (Davide's) or new one (Giovanni's)
  const bool enforce_old=false;
  const bool enforce_new=false;
to enforce old/new version.