Implement double-double and quad-double arithmetic under a permissive license
For very ill-conditioned problems, higher than IEEE double-precision arithmetic can be useful for debugging. The de facto standard implementations of double-double and quad-double arithmetic, which essentially concatenate the mantissa of two or four doubles, is in the QD library of Hida et al.
A modern, templated implementation under a permissive license would be needed before integration into Catamari. The official LBNL distribution page of QD has even been shut down.