I think perhaps the way I would state the subset problem is that white keys are in a way the "subset that sounds good" to the culture where the piano keyboard arose. More in the vein of discussion than suggesting you'd need to change anything :)
The black keys are means of transposing that subset.
It is a very interesting but in a way unrelated insight that the black keys also form a consonant group.
And then music and musicians naturally coevolve with instruments and do whatever they please!
This appears to differ from the Janko layout, though they both apparently share the feature that "if you know the shape of a particular chord or scale, you can automatically play the same thing in another key just by moving your hand" (so long as you have five rows of buttons, in the accordion layout.)
There are many ways to skin this cat apparently, though as with QWERTY, established convention is hard to change.
Changing keys on standard keyboards just seems unnecessarily difficult.
My primary instrument is piano so it's definitely involved rewiring my concept of which notes are 'close' and 'far' to match the circle of fifths. Until now I never noticed how in addition to the usual cadences, many tunes have further rising/falling sequences of 5ths in the bass.
You're right the harmonic model used is a bit of a bodge. I didn't know that about hammer position, very interesting. Let me know if you ever get around to trying different timbres in the code.
I think we are more in agreement than you think on clustering, though. Although I neglected to discuss what's anatomically reachable, and also the nuances of history, my point is that the desired subset you mention can be defined by using clustering to pick a subset that sounds good.