Don't throw in sparse LU for singular matrices. Fixes #13781.

[andreasnoack]

This is similar to what we do in the dense case and it avoids the error when trying to compute the determinant of a singular sparse matrix. I've also changed the exception thrown when trying to solve with a singular matrix to SingularException, but here I set the numrical value to zero. In the dense case, the number is the smallest zero pivot index.

[tkelman]

Does umfpack provide the first zero pivot information?

[andreasnoack]

I've read the documentation in the header files as well as looked at the source for the factorizations and he doesn't seem to keep track of the location of the zero pivots - only the total count. We can of course compute it from U, but it's quite some overhead for a calculating a number that very few cares about. It would also require some restructuring of the error reporting so I'm not sure it's worth the trouble.