FWIW pbrt-v4's "imgtool" utility has a "makeemitters" option that takes an image and emits a quad for each one, though it uses the bilinear patch primitive that is in pbrt-v4 and not pbrt-v3. That would be easy to update to generate 2 triangles for pbrt-v3. pbrt-v4 also has built-in support for the use of textures to describe emission (not yet documented, so you'd have to dig through the area light's parameters...)
The colored glass panel with a uniform light behind it is a clever idea (and after all, that's similar to how displays work these days anyway.) One problem with that is that it will thwart pbrt's direct light sampling algorithms, so light will only be found when an indirect ray randomly happens to hit the glass and then refract to the light behind it.
It isn't clear to me why the reflection would be blurry like that. It's a mirror with perfect specular reflection, or does it have non-zero roughness?
Thanks,
Matt