Let me be clear, I'm assuming the anti-aliasing filters are sufficient to get a good digital signal conversion, and that the filter is used before resampling down to 28.4 MHz for storage.
The harmonics don't contain information and thus we only need to sample at the highest frequency deviation for the FM signal. A good ADC will reduce aliasing from the harmonics. Clipping is a normal stage in FM systems. In this case it's just done at the transmitter stage instead of at the receiver.
If we want to preserve the disc signal itself (i.e. the pulses), and not just the composite signal, we'd want to sample at at least twice the bandwidth of the optical detector.
I've heard that the SNR is around 50 dB for video and 110 dB for audio. This is regarding the demodulated signals however. For FM, we will only need 1 bit per signals. The dynamic range is only dependent on the frequency deviation.
With all this said we have some choices:
1) Store the PWM signal This is the most similar to how the disc is stored. It requires a lot of bandwidth, and the emulator has to demodulate everything.
2) Store the FM signal This is essentially the same as above but I don't think it can easily recreate the original medium. It requires less bandwidth though.
3) Store the streams This demodulates the audio and video but doesn't process the video signal. The emulator has to do it.
4) Capture the video This demodulates everything and also converts the video to some lossless format. You also have to define a format for v-blank data.