Function maxpool2d_raw 

pub fn maxpool2d_raw(
    input: &[f64],
    out: &mut [f64],
    n: usize,
    c: usize,
    h_in: usize,
    w_in: usize,
    ph: usize,
    pw: usize,
)

2D max-pooling — NCHW layout, stride = pool_size (non-overlapping).

• input: [N, C, H_in, W_in]
• out: [N, C, H_in/ph, W_in/pw] (floor division, pre-allocated)

All index arithmetic uses u64 to support large spatial extents.