By all means, educate me on how non-linearity can make hair cells responsive to frequencies > 20kHz.
1) you have two frequencies at 23Khz and 25khz. Say they mix vith a quadratic non-linearity then you get a frequency at 2Khz
2) this is not simply aliasing
3) The case of 2 frequencies is degenerate but if I have multiple simultaeous frequencies mixing down below 20Khz then I can reconstruct the original spectrum under commonly true assumptions (such as sparsity).
There can also be non-linearities in time and phase as well. A trivial example is that a loud sound a moment before can block hearing a quiet sound a moment later.
Finally imagine how the ear actually works. It's not just a comb spectrum analyzer. It's actually sensitive at frequencies in between discrete values. Moreover in order for sound to reach the folicles the furthest in it has to pass by the folicals at the front, Therefore there is a great deal of mixing in time and frequency that can occur from this design.
I'm not saying this matters or that 192 K solves it if it does matter. I'm just saying the linear analyses assume their own conclusion and parade out Nyquists theorem as though it applied.