Function compute_block_l2_errors 

pub fn compute_block_l2_errors(
    original: [&[f32]; 3],
    reconstructed: [&[f32]; 3],
    mask1x1: &[f32],
    xsize_blocks: usize,
    ysize_blocks: usize,
) -> Vec<f32>

Compute per-block masked weighted L2 error between original and reconstructed XYB planes.

Each block’s error = sum over 8x8 pixels of: mask[px]^2 * (w_x * dx^2 + w_y * dy^2 + w_b * db^2)

All planes and mask have stride = xsize_blocks * 8.