Hey team,

I'm very new to Rust, and very new to using the Image-rs crate -- so please forgive the potential mistakes I could write/say.

I'm trying to add a block enumerator for incoherent ImagBuffer access.
This is usually faster than regular pixel enum when we want to perform a column-major operation on a image, even though the memory layout is row-major.
Basically the following code:

fntranspose_enum<P,Container>(input:&ImageBuffer<P,Container>,output:&mutImageBuffer<P,Container>)whereP:Pixel,Container:Deref<Target =[P::Subpixel]> +DerefMut{for(x, y, pix)in input.enumerate_pixels(){       output.put_pixel(y, x,*pix);}}

can be outperformed by the PR modified version:

fntranspose_enum_blocks<P,Container>(input:&ImageBuffer<P,Container>,output:&mutImageBuffer<P,Container>,block_size:u32)whereP:Pixel,Container:Deref<Target =[P::Subpixel]> +DerefMut{for(x, y, pix)in input.enumerate_pixels_blocks(block_size){        output.put_pixel(y, x,*pix);}}

On my old 2016 laptop I get between 1.3x to 2x perf improvements for some adequate block sizes when tranposing 2K - 4K and 8K images. I still didn't properly becnhmarked on my desktop machine though (Ryzen 7 5800x) but will soon do.

Let me know if that's of interest for the image crate, and please let me know of any dumb mistake I could have done in the code. Note that I randomly chose function and variable names, but we can completely change that, and I can also make a mutable version of the enumerator.

I license past and future contributions under the dual MIT/Apache-2.0 license,
allowing licensees to choose either at their option.