We could probably increase the robustness of our chi2 backend if we used a high-quality linear solver like Eigen (https://libeigen.gitlab.io/eigen/docs-nightly/group__TutorialLinearAlgebra.html). Our current custom implementation is fast, but probably a little too eager to declare a matrix singular. We originally tried LAPACK, but that had high overhead for the small matrices that appear in LS. Eigen's solver probably has less overhead, since it doesn't involve a call to a shared library.
x-ref #85