[Lapack] Bugfix for inclusion in upstream version

[Wolfgang Dobler]

I have a bug fix for lapack that I think should go into the next version.

The bug concerns error handling and leads to division by zero and hanging
of the call to dgelsd().

  See http://icl.cs.utk.edu/lapack-forum/viewtopic.php?p=1747


W o l f g a n g

PS: I would really appreciate to hear _anything_ in reply to this message
  that would prove that lapack maintainers do exist, so we (and others)
  don't have to switch to a different library in the near future.