Function product 

pub(crate) fn product<R: Send>(
    m: usize,
    n: usize,
    k: usize,
    op: impl FnOnce(Par) -> R + Send,
) -> R

Run an m × k by k × n faer product with the treams-rs budget.

Matrix products split their output among workers and give the same bits at every budget. A matrix-vector product (m or n equal to one) splits its inner dimension instead, which changes the last bits with the worker count; it is bound by memory bandwidth and stays sequential.