Function matmul_raw 

pub fn matmul_raw(
    a: &[f64],
    b: &[f64],
    c: &mut [f64],
    m: usize,
    k: usize,
    n: usize,
)

Matrix multiply: C[m,n] = A[m,k] × B[k,n] with Kahan-summed dots.

a, b are row-major contiguous slices; c is the output buffer (must be pre-allocated to m * n).

Uses in-place KahanAccumulatorF64 — zero heap allocation per dot product.