move the icc shader to the rest of KWin's OpenGL code, and add an autotest for it

The test just uses the ICC profile I created from my laptop, transforms an sRGB image to be fitting for that display and compares the result from lcms vs. KWin.

I can't say I'm 100% happy with the result, but as even lcms can slightly disagree with lcms, depending on the options you give its transform, it's not too bad. The max. error we have is most likely from linearly sampling the 3D lut; I don't know exactly what calculations the GPU drivers use there, but it should help to manually do tetrahedal interpolation in the shader instead. Some error likely also comes from slight differences in how matrices are calculated, where using doubles instead of float might help, but that can be investigated later.