In essence, you cannot rely on the engine to generate lightmaps.
Not for complex model, and not for simple oness too since it doeant take into account texel density and a bunch of other things like power of 2 pixel scaling.
Take the model to something else - other autodesk stuff probably?
Or Blender.
From there, you slice up the model into parts that make sense, and generate lightmaps for it manually.
Possibly on a base power of 2 textures that is around 1024px^2 and no larger.
You align each UV island vertex to the center of the pixel it sits on, so as to have clean bleeding when you mip the texture up and down.
The end result at that point is a model with a set of UVs that works across anything you push them onto. Regardless of the size you pick when baking.
Remebering that 1024 is probably the minimum size you want for lighting anyway.
The more pieces you separate out as different meshes, the better quality of a light bake you end up getting.
But as you do, you need to keep track of the texel density of the UV map, and always keep the pieces at the same scale, so that the light bake process doesn’t jump in quality from one object to another, but is consistent acrosd the board.