There are multiple things happening in the algorithms presented in the link you shared, but the most important thing that's happening is not possible in Blender's compositor.
- Converting an image into a fixed of points
- transfering that set of points as a parameter into a hypothetical "voronoi" node. There is no such data type
- using a vornoi algorithm to fill in polygons with sampled colors from a set of points
Now, in theory you could reduce that set of instructions into just one step
but to do that is no small chore
There are plenty of clues in the codegolf post on how to process the image, how to find edges, how to make that bitmap into a set of points, how to process the points into polygons, and how to fill them.
I'd like to offer up the code for a python compositor node that just does all that for you, but it would just take a terrible amount of time. An onerous amount of time. Enough time that the short answer to your question is "no."
On the other hand! If you're also interested in exploring the general effect and breaking out of the compositor-node-constraint, there are other things you could explore. (in your comments, it seems you're only interested in voronoi specifically, but i'll try anyway)
Here's an image that was made with a "lens" made from a box, one side heavily subdivided, displaced with a cloud texture, and then decimated. The lens is using a glass texture with 0.02 roughness to generalize the colors coming through each face. If the displace filter was made to transpose over time, the facets would change and shimmer.
While it's not exactly what you asked for, in my opinion, it falls into the non-photo-realistic category. I think you could go quite far with it -- especially with render layers feeding into the compositor.