blas error and crash when doing ldivide/rdivide of a scalar by a vector

[Jason Riedy]

And John W. Eaton writes:
>
> How?  We call DGELSD with lwork == -1 and ask it to tell us how much
> work space to use.

Yeah, it's more painful than I thought, especially since some
people (well, me) switch LAPACK libraries at run time.

And worse, the documentation in {C,Z}GELSD is completely busted
so I can't claim that some obscure formula *must* work.  Ugh.
The problems stretch way back; I don't know how they lurked
unnoticed.  There may be no sane Octave-level workaround, but
I'll get it fixed in LAPACK before the next release (whenever
that is, no control over that part).

Jason