Gurobi reports PB instances as satisfiable when they are not, large coefficients

[OrestisLomis]

I attached some instances in the issue where Gurobi reports them as satisfiable, but when checking the constraints there are certain constraints with very large coefficients that are in fact violated. Perhaps a tolerance issue?

unsat-instances-gurobi.zip

[hbierlee]

As discussed internally, this is indeed a tolerance issue also encountered by @ThomSerg, who found using IntegralityFocus resolves it, but it'd be interesting to see whether adopting the new expression-based API from Gurobi could help as well. This issue is a good starting point for that PR.