Because it's impractical, unnecessary and wasteful. There are a number of processes that by their nature require global knowledge. These are not the sorts of processes that need to be done in the client (many of them in fact are done up front precisely so that a client will not need to do them), but they are the sorts of processes that a developer will need to do if they are to be able to make a change and see its effect. Also, there are good reasons to generate assets at a higher resolution than needed in the final artifact.
Just like if you're rendering a 3d animated film you might need a more powerful machine than those who are only watching it, or if you're training a neural net, you need a more powerful system than if you're merely running it.