Given a directed "follows" graph as a Python dict[str, list[str]], implement recommend_two_hop(graph, user) that returns the set (or a sorted list) of accounts followed by the user’s followees that the user does not already follow, excluding the user themself. Deduplicate recommendations; if you return a list, sort by descending frequency among two-hop neighbors, then lexicographically. Example: graph = {"A": ["B","C"], "B": ["C","D"], "C": ["E"]} ⇒ recommend_two_hop(graph, "A") = {"D","E"}.