Gurobi will always linearize your constraints provided in the OPB format. There is no parameter to apply the propagation described in the paper you mentioned. Thus, in order to test linearization vs. propagation for your model, you could refer to the open-source solver SCIP, which handles the non-linear terms as a special $$\texttt{and}$$ object. If you conclude that linearization works best, you can then still switch back to Gurobi.