find_best_packing(): optimize required order storage to 2 * std::size(subjects)

[lyorig]

Implements #16.

As it stands, this PR is a rough draft cobbled together in my spare time that compiles and seems to work. There are most likely some things to improve (especially variable names), but hopefully it'll serve as a good starting point.

[lyorig]

CI seems to have caught something weird. I'll try to look into it soon™.

[lyorig]

Oh yeah, I got some stuff completely backwards. Will try to fix if time allows for it.

[lyorig]

Checks are currently failing due to a single misplaced rectangle, which is interesting. I'm still pretty short on time, so I won't get into debugging it right away—but it may very well be caused by some trivial UB or other coding error on my part (although ASan + UBSan stayed silent).

[lyorig]

@geneotech, I've decided to rear my ugly head again, and hopefully finish up this PR after God knows how long. The output mismatch seems to be caused by individual sorts being influenced by the result of the previous one; to fix this, I instead chose to allocate 3 * std::size(subjects) pointers, and copying the initial state to the "staging area" after every sort.

So, tests are currently passing, but I wonder whether the previous iteration (which did not pass tests) was still correct in the algorithmic sense of the word, just producing different (but still valid) results. You'll probably have to help me out here with your expertise.

[geneotech]

Hey, since we confirmed the tests are green after introducing the third backup array to sort from, I believe it's okay if we just have 2 buffers to not overcomplicate this, and just change the test results. The sort is likely just non-deterministic when there's e.g. two rects with the same area.

[lyorig]

Understood, I've reverted the changes and changed the test output accordingly.