You don't actually need the NCS relationships in order to refine with NCS restraints because the default is to use torsion-angle restraints which only use local relationships, not overall NCS matrices.
So you can go ahead and use your rigid-body-refined starting model with torsion-angle restraints in phenix.real_space_refine. That will not impose perfect helical restraints, but I am guessing that your map already has near-perfect helical symmetry so that should not matter very much.
Also you might first run with just one monomer which will be much quicker and give you almost the same answer.
Let us know if that does not do it!
All the best,
Tom T