Note about #1
The transformer have (depending on used type) a limited bandwidth that may significantly change the curve form you measure on. But your design with a rectifier and capacitor means that you only measure the peak voltages. There is a fixed translation between peak voltage and true rms voltage as long as you have a pure sine wave. As soon as the output starts to clip, that relation fails.

It would have been way better if you had been able to sample the ADC output of the transformer at high enough speed to let you compute the RMS energy instead of measuring the top voltage.

Also, the capacitor means that the design is a low-pass filter, slowing down your regulation loop because the capacitor isn't instantly emptied if the primary voltage to the transformer quickly drops.

#2
Peak voltage (as you have already noted) measurement does not work if you have distortion.

But even more importantly is that your program should be written so that it never fails to measure. If you can't guarantee real-time performance, then you can't guarantee that you will not create big disasters on the high-power side. Your regulation loop just has to be fast enough. It's not an option but a requirement.

#3
I'm not sure what you mean with ZC, and feel you have to give a better description of exactly what you are trying to do.

Right now, you are busy measuring absolute values. In some situations, it may be way easier to measure a delta, i.e. the residual error. This can be done by scaling the output voltage and then subtract a pure sine wave of correct amplitude. If the output voltage is too high or low, you will get a positive or negative voltage to measure. If the output voltage starts to clip, you will also get an error difference that the processor can respond to.

From the low number of samples you seem to take in the feedback loop, I guess you are not trying to waveform-adjust the output - you are just trying to figure out if the output voltage is too high or low?