The bigger challenge here is pixel architecture, but if apple is actually slicing up wafers into a couple million pieces to build these displays, they are already sort of moving away from the typical TFT architecture and may be able to integrate more complex pixel drivers, potentially including things like touch sensors directly onto the pixels.
I don’t understand what you mean with the GPU. It is has no information about the exact color of the LED.
The shift in wavelength is primarily determined by temperature and current, and they work in opposite directions so sort of cancel each other out. And in any case, we're talking about well-characterized shifts on the order of a few nm over the operating range. The eye's cones are broadband, so you're not going to notice wavelength shifts, especially compared to the brightness variations over the same range.
This is a big deal for white LEDs because you have no control of the resulting color temperature (the phosphor emission and blue component wholly determine the output), but for an RGB structure, you have pixel-level control over each component.
As to binning LEDs that works because it is constant. You can calibrate it once and done. But if you change the brightness by changing the current, it means your calibration is out of wack. Perhaps you can make a calibration at multiple current settings, but that seems inconvenient when using PWM will achieve the same thing.
So if the color shift is noticeable enough to require correction, then it's definitely enough to substantially decrease the color gamut as well. And so a range of wider-gamut colors simply can't be compensated for at all.