Consistency when assessing and correcting KKT errors after IPM and PDLP

[jajhall]

After IPM without crossover, when optimality is claimed using the relative primal/dual infeasibility measures, there can be absolute primal and dual infeasibilities. These are removed by truncating the smaller of the primal and dual infeasibility, leaving no KKT errors

After PDLP, KKT errors are simply measured, and optimality is lost.

In this respect IPM and PDLP should operate the same way.

After both, the primal and dual residuals should be computed, applying corrections and possibly losing primal/dual feasibility

If a primal variable is off its bound by more than the feasibility tolerance (even marginally) the absolute value of its dual value is considered to be a dual infeasibility. This seems excessive.