Build a Friend Recommender

You are given a simple social-graph codebase where each user has only an `id` and a collection of current friends. Starting from existing functions and unit tests, complete the following tasks: 1. Fix `valid_recommend(user, candidates)` so it rejects invalid recommendations, including recommending the user themself or someone who is already a friend. 2. Implement `random_recommend(user, candidates)` to return a random valid recommendation. 3. Define a measurable recommendation-quality metric using only graph structure, since no profile features are available. 4. Implement a `mutual_friends_recommend` strategy that ranks non-friends by the number of mutual friends, and add tests for it. Return any deterministic tie-broken result when multiple candidates have the same score.