Parallelize Utils::build_symmetric_csr

[DanecLacey]

The Utils::build_symmetric_csr() routine takes a large portion of the runtime with sizeable matrices. Since this contributes significantly to preprocessing time in some cases, having a good parallel scheme is cruicial.

[DanecLacey]

Done, copied from https://github.com/RRZE-HPC/RACE. It seems that nonzeros within a row are not deterministic. I could sort them, but I think for now it is okay.

[DanecLacey]

At least for the unit tests, we should sort nonzeros by row. That way, we at least know that the correct nonzeros are on each row.