Function fold_right_lerp

pub fn fold_right_lerp<P, PE>(
    evals: &[P],
    log_evals_size: usize,
    lerp_query: PE::Scalar,
    out: &mut [PE],
) -> Result<(), Error>
where
    P: PackedField,
    PE: PackedField<Scalar: ExtensionField<P::Scalar>>,

Specialized implementation for a single parameter right fold using linear interpolation instead of tensor expansion resulting in a single multiplication instead of two: f(r||w) = r * (f(1||w) - f(0||w)) + f(0||w).

The same approach may be generalized to higher variable counts, with diminishing returns.

Please note that this method is single threaded. Currently we always have some parallelism above this level, so it’s not a problem. Having no parallelism inside allows us to use more efficient optimizations for special cases. If we ever need a parallel version of this function, we can implement it separately.