This will sightly depend on how confident you are about the binding mode of your complex. It seems that you already know for sure what residues will interact and how, right?
If this is the case, and you do not want to sample much your solutions, then you might want to go for unambiguous restraints between the residues atoms and the ligand ones.
If you have some uncertainty, then any residue interacting with the ligand could be added to the list of active residues (and then generate ambiguous restraints this time.) HADDOCK will model, hopefully correctly, the potential H-bonds.
Anyway, several runs with different setups can be done. Then your understanding of the system will certainly make you decide which ones are the most relevant.
Hope this will help,