Have you tried building the star mask on linear data with only a small Midtones stretch? The true stars will generally be brighter. This won't get you all the smallest stars, unfortunately, but it might be good enough to mask all the big ones without including the galaxies.
Cheers,
Rick.