The answer is quite simple. We don't do either of those. 

Our approach is that we don't actually store a trillion or so stars, we instead generate them procedurally on the fly. Stars that don't need to render will be removed and stars that are about to come into view will be generated. The scope of this calculation can be adjusted in one of the behaviors in the asset, as it may depend on your star material, shader settings, spatial distribution/subdivision and so  on. Since all star generation is based on a seeded random generation coupled with distribution functions, the result will always be the same for a given quantity/size/distribution setting of the galaxy and the accompanying star fields. .