Function cpu_batch_step_f64 

pub fn cpu_batch_step_f64<A, S, F>(store: &S, kernel: F) -> AccelStepResult
where
    A: SoaExtractableF64,
    S: AgentStore<A>,
    F: FnMut(&mut [Vec<f64>], usize),

CPU-side batch step over f64 SoA columns.

Parallel to cpu_batch_step but preserves double precision end-to-end. Use this when f32 would introduce unacceptable rounding — e.g. long-horizon integrators, stiff dynamics, or scientific workloads.

kernel receives (columns, n) where each columns[c] is a mutable Vec<f64> of length n.