IMO the little kernels and splines always have been a red herring.
At worst they have absolutely nothing to do with this.
At best they refer to the highest-rate anti-alias filters that MQA uses. But given how crude these filters are, their actual shapes are of little importance.
'If you can't dazzle them with brilliance, baffle them with "science".'
The pulses are stand-ins for real signals. And of course we want to locate real signals temporaly with whatever accuracy is required. But this does not pose any exotic demands on the detecting apparatus, be it a machine or your auditory system.
An example from my daily line of business: a spacecrafts's attitude is determined relative to the fixed stars. This is done with a camera and a slightly defocused lens. The camera may be coarse, e.g. a measly 512 x 512 pixels, and yet star positions are determined to deep sub-pixel accuracy, translating in mere arc-seconds attitude error. Why? It is not the sampling that is the limiting factor (the pixel array), but rather the system's noise floor. Sounds familiar?
Dirac pulses are mathematical abstractions: infinitely short, unit energy, hence infinite amplitude. With some hyperbolic analogy they are quickly likened to nuclear explosions or the big bang, both being wrong, but they convey the idea that a Dirac has no place in music.