ignoring the aspect of "what light goes where" you first need to actually define colors. the obvious options are blackbody radiation (blender already has color temperature) and a CIEXYZ thingy instead of a color wheel. if you're doing, like, computer screens specifically you could have an rgb gamut I guess.

but if you're doing some serious physical modeling you'd want spectral lines. especially for absorption and shit. and at that point you can fuck the cie color space and just start putting points on, like, an EQ curve but for tthe color spectrum. I guess working with color as a spectrum of intensities would make it much simpler. like, transluscent objects would just be, essentially, multiplying.

and as for the diffraction, idk, maybe you could just have light sources sending out a few samples for spectral colors and then just... interpolate the full spectrum between those somehow. not unrealistic if you can somehow bake it to specific objects/lights