Anyone know realistic MiM capacitor leakage? Models seem to say it's exactly 0. Is it so low that I can ignore it? I don't think the device details provide the necessary info to calculate it.

Add me to the reliable sources that say it's small enough to ignore. However, MiM caps are sensitive to the antenna effect, and you are going to have to couple both sides of it to diffusion somewhere. There will be leakage through the diffusion diodes. The only perfectly (or close enough to it) insulated capacitor in the process is the SONOS gate (if you, or anyone else, can figure out how to program it). That, of course, is mostly just repeating what Luis already said.

Oh no my hopes were so high

Even if you don't connect to an antenna diode, you'll be connecting to diffusion somewhere in the form of a FET drain or source, and that's a piece of diffusion that makes a p-n junction to either well or substrate, which is a reverse-biased diode that leaks.