For problem F, I'm not really sure why you needed both persistence and a 2D data structure; I just statically built up a 2D data structure representing the pre-flip rectangles, and then used its nodes as my graph nodes. For the 2D DS, I recommend static (offline) segtree of segtrees. For each outer segtree node, gather the set of all inner coordinates which will be updated/queried at this node and then coordinate compress to build an appropriately sized inner segtree. It uses O(n log n) memory and has O(log^2 n) query time, and also lets you just use segtree book code if you want. The static outer segtree is also easier to use than persistence.

For E, I did expected implementation trouble in the beginning, but in the end it turned out to be rather straightforward. If you take the Voronoi cells with half-plane intersection, the rectangular boundary doesn't make the problem harder, you just put all adjacent vertices in Delaunay triangulation with four more halfplanes. Then you can get a polygonal representation which transforms the problem into pure graph problem (although with some real-number hassle)

the problem is that he uses windows 10.