After the binary search converges, also simulate predicted rating at
required±1 average. Display the three rows in the modal so the user can
see how sharp the requirement is — e.g. whether averaging 1 point lower
costs them 1 point of predicted rating or 5.
When a player has rating_history (graph) but no round_history (per-round
detail), calculating a target produced a dead-end error. Now the modal
detects the NO_ROUNDS case and shows a button that triggers the existing
refresh-round-history endpoint and re-runs the calculation on success.
Handles the 24h rate-limit and other refresh errors explicitly.