Improve efficiency of rotate_imgs

Currently, rotate_imgs is not fully utilizing the CPU due to the way in which the images are divided amongst the workers - each worker processes one image at a time, which is not efficient given the small image size.

This can be solved by passing a larger number of images to each worker (rather than one at a time, pass e.g. 10). More formally, for n cpus and m images, pass floor(m/(n-1)) images to the first (n-1) cpus, then pass m-floor(m/(n-1)) to the last.