How do you calculate if they should be lit or not? And to what degree? That requires megamaths, and that's what GPUs are. There's very little room for coprocessors any more because the workloads aren't specialised and trivial. That's why we have unified shaders, for example. Also, any lighting hardware would be locked to a specific type of lighting, unless you make it programmable at which point it becomes a generic compute unit. I suppose a Global Illumination unit would make sense as pretty much all games can benefit, even toon rendered. Although any dedicated silicon can't be repurposed, which is why the Holy Grail is unified processing power and a single maths engine doing whatever you want, from 3D graphics to audio simulation.Imagine something like that today, but deciding if the pixels to the screen are lighted or not.
One thing I have considered is actual 2D hardware. Currently 2D is handled in 3D with quads and it seems terribly wasteful. If I want custom characters, I have to write character pieces to the character texture which is dog slow (at least in Unity!), or use a draw call for each piece of a character. In the 16 bit era I'd just copy arbitrary bits of memory for each part to the destination location. Blitting 2D has a lot of value still, I think, especially in mobile. I don't know if a 2D blitting engine could be written in GPU nowadays?
