Add slacks in the barrier reformulation, not in the model

[cvanaret]

TODO:

• moved slacks from HomogeneousEqualityConstrainedModel to PrimalDualInteriorPointProblem
• phase transitions
• correct residuals (stat and cmp from slacked model, constraint violation from original model)
• residual scaling
• fix remaining failures on infeasible instances

Consequences of the refactoring:

• keep track of structure in a finer way (identity block in inequality Jacobian, regularize only original variables)
• it's cheaper (fewer variables) to do l1 relaxation then slacks, than slacks then l1 relaxation