// TODO: Himada + dfdx integration
// dfdx's type-level tensor shapes require const-generic gymnastics.
// For now, use the C ABI or Dispatch directly when working with dfdx:
//
// use himada_dispatch::*;
// let d = Dispatch::<DotKernel>::new("dfdx_dot", vec![]);
// let result = d.compute(&a, &b);
//
// See docs/c/README.md for C ABI setup.