Thanks again for your reply. I found the issue and I will say it here, it may help other people in the future:

the vertical pressure gradient was actually ok because gravity is activated in my model. At such low velocity, rho g h becomes important since the inertial terms are weak. the solver was detecting the vertical pressure gradient due to gravity correctly but it was forced to assign a 0 Pa pressure all over the outlet which was causing the backflows. The solution to that is activating the "average pressure specification" option in the pressure outlet panel. This will make sure that the area-weighted average of pressure on the outlet plane is zero, while it gives the solver the flexibility to change the pressure in different cells.