You should indeed NOT do (1) and (2) since those calculated restraints are incompatible with the manual active/passive residues you have added. They do not appear in the 2nd docking run protocol anyway. You might be mixing the two protocols there.
So you can apply (3), this will have the described effect, but applied to your restraints (any active/passive definition are ambiguous restraints). As explained at the beginning of the section Setting up a new docking run targeting the identified binding pocket this will allow to first target the binding site and make sure to draw the ligand in, then leave it enough freedom to explore several conformations.
But to stay inline with the protocol proposed, it would be maybe nicer to create, with the GenTBL interface, two sets of active/active and active/passive residues and apply them the same way. You can do it with your own active residues, as it is described in the tutorial (same section).
Hope this will help,