multidimensional optimization

[Brian Gough]

This is done now -- see the documentation for details.

Brian Gough writes:
 > I'll provide an alternate method setf which accepts precomputed
 > function values. That will be the simplest way to handle this case.
 > 
 > Also I'll move the function values (flower,fmin,fupper) into the
 > algorithm-independent state, so they are more accessible (e.g. so that
 > the value at the minimum does not have to be recomputed).