Next-pointer and quad tree serializers follow a node that links back to itself #428

#388 gives the binary tree serializers a visited check, but the next-pointer and quad tree serializers walk their children with none, as the review of #388 pointed out. In Org Chart Row Navigation, a linkRowNeighbors that sets root.left = root and leaves root.next null passes the next-pointer check on every level, so the breadth-first walk never ends. In Terrain Mask Tiling, a buildTileTree that returns an internal node whose topLeft is itself makes the preorder walk recurse until the stack runs out. Seven of the ten serializers below then report no case at all, which is the failure #386 reported for binary trees, and the other three blame the stack rather than the tree.

Measured on 3198a99 (macOS 26.5). The page rows come from the interview page in offline mode, the compiled rows from the harness built with cc, g++ and javac and run with a 10 s timeout.

Proposed change: the next-pointer serializers refuse a node they reach a second time with "Returned tree must not reach a node twice.", reported as the case's error and keyed on the node rather than its value, as #388 does for binary trees. A correct answer there is the input tree with its next pointers set, so no valid answer reaches a node twice. The quad serializers should refuse only a node that is its own ancestor, tracked along the recursion, because an answer may reuse one leaf for every uniform quarter: a g++ answer that builds one 0 leaf and one 1 leaf and returns them from every uniform region passes all five Terrain Mask Tiling cases on 3198a99, and a visited set would refuse its checkerboard and mixed cases. Since that check can tell a cycle from a shared leaf, its message can say "Returned tree must not contain a cycle." Each serializer gets a self-linked case that fails before the change, plus cases that must keep printing: distinct nodes with equal values in both, and a shared leaf in the quad serializers.