Check elision of bounds checking for simple loops

When checking unsafe usage for https://github.com/image-rs/imageproc/issues/371 I found a few cases where benchmarks showed a speedup from using unsafe getters despite the accesses having the following form:

    for y in 0..image.height() {
        for x in 0..image.width() {
            ...
            let p = image.unsafe_get_pixel(x, y);
            ...
        }
    }

Check the generated assembly from using the current accessor functions, starting from a trivial loop, e.g.

/// .
pub fn trivial_loop(image: &mut GrayImage, d: u8) {
    for y in 0..image.height() {
        for x in 0..image.width() {
            let p = image.get_pixel(x, y)[0];
            image.put_pixel(x, y, Luma([p + d]));
        }
    }
}

/// .
pub fn trivial_loop_unsafe(image: &mut GrayImage, d: u8) {
    unsafe {
        for y in 0..image.height() {
            for x in 0..image.width() {
                let p = image.unsafe_get_pixel(x, y)[0];
                image.unsafe_put_pixel(x, y, Luma([p + d]));
            }
        }
    }
}

Start with the current accessor functions used by the trivial_loop and trivial_loop_unsafe examples in the issue. Generate and compare their assembly, beginning with the nested height/width loops, and determine whether safe accesses retain bounds checks. Done means documenting the comparison and identifying whether the safe form can match the unsafe form.