Function run 

pub fn run(
    client: &mut RespClient,
    old_cmd: &[Vec<u8>],
    new_cmd: &[Vec<u8>],
    old_shape: Shape,
    new_shape: Shape,
    samples: u64,
) -> Result<ShadowReport>

Run both commands samples times and compare each pair.

Both sides are issued on the same connection, back to back, so the window between them is as small as this can make it. A row written between the two reads shows up as a divergence, which is why a single disagreement is a lead rather than a verdict — the report carries the count so a rate can be read off it.