Function pack_chw_into 

pub fn pack_chw_into<T: PixelComponent, const CHANNELS: usize>(
    input: ImageView<'_, T, CHANNELS>,
    scale: f32,
    mean: [f32; CHANNELS],
    std: [f32; CHANNELS],
    output: &mut [f32],
) -> VisionResult<()>

Packs and normalizes interleaved pixels into a reusable planar CHW slice.

Large multi-channel inputs dispatch across channels with safe scoped threads. Smaller inputs remain scalar to avoid scheduling overhead.