This is not a direct answer because video games, any real time renderer in general ( not sketchfab yet ) uses LOD ( Level of detail ) So every asset , not a background asset, that has the possibility to be close and far from the camera at different moments in the game has multiple versions of geometry. So if a asset is very close to the camera they load a high poly count version of the asset , LOD 0 , as it moves further away they load the assets with less geometry. Sometimes such assets are all tailor made but many times they use software to create the LOD versions , stuff like simplygon or in house systems. I dont know the exact numbers but im pretty sure the LOD 0 , or highest poly version would be quite heavy 200-300k maybe?
Anyway as a artist whether you are doing cg models or video or image design you always do your work at the top most quality. This is your “master” file. As it is easy to then create the lower versions ( as mentioned this is sometimes even automated ) If you work on the low res version first then it is not possible to up res your stuff if needed in a easy manner, or not at all. Think of images and videos… you can downscale and not lose quality … but you cant upscale without quality loss.
Anyway I fully understand your workflow , i too also create a initial version to block out shapes etc , and then do the body again afterwards.
looking forward to some more progress.