Alternative seam carving implementation

Hello !
This is not an issue, but a follow-up on: https://github.com/image-rs/image/issues/982#issuecomment-508858512

I finished implementing what I wanted to, and my results are quite disappointing, but I wanted to keep you updated nonetheless, since you asked :)

So here are the results:

This represents the time taken versus the number of seams to remove. "this crate" is seamcarving

My approach, that I thought was smart, seems to be orders of magnitude slower than the traditional dynamic programming approach taken in imageproc.

The idea was that by using a path finding algorithm, only part of the image would have to be visited, and the higher energy regions would be completely ignored, without even having to compute the energy at every position in them.

I haven't investigated the performance issue yet, but it seems that the added complexity of path finding far outweighs the gains.

Read the linked issue discussion and compare the seamcarving crate's path-finding approach with imageproc's traditional dynamic-programming implementation. The report identifies a performance gap but does not name repository files, tests, a concrete change, or acceptance criteria, so the intended definition of done remains unspecified.