Implement superlu for scalar valued matrices

There is no reason why it should be a C-pointer.

Previously, only BCRSMatrix<FieldMatrix<T,1,1> > was tested.

Previously, all code was in the specialization of that class
for BCRSMatrix<FieldMatrix>.  That was overly restrictive:
most of the code does not actually rely on the precise type
of the matrix.  As a first step towards allowing more general
matrix type we move all the code into the general class.

In the debug output, 'n' was used for the number of matrix columns.
However, 'n' is the number of columns of the FieldMatrix objects
that serve as matrix entries.