From conversation with @lgottwald

[jajhall]

• When HiGHS finds that an LP is infeasible in the scaled space it should check in the unscaled space
• When optimal in scaled space, but not in unscaled space, HiGHS should resolve automatically with tightened tolerances
• HiGHS' primal/dual rays can be inaccurate - try using iterative refinement with quad precision residuals in their computation
• Should be possible to use HiGHS numerical linear algebra to identify rank of the equations in the LP. HiGHS would have to be given a rectangular matrix to factorize. That's not something that it could do at the moment, but probably not too hard to add the code required.

[jajhall]

Done or WIP