Function chunk_writes 

pub(crate) fn chunk_writes<B: Backend>(
    xhat_blchp: F<B, 5>,
    bhat_blchr: F<B, 5>,
) -> F<B, 5>

The per-token write W_j = Σ_c x̂[j,c] ⊗ b̂[j,c] ([b, h, j, p, r]) — used by the backward’s da gradient.