Implement Peer Mentor Stats Aggregator

Prefer a Supabase RPC function (PostgreSQL stored procedure) for the core aggregation rather than multiple separate queries — this reduces round-trips and leverages the database's query planner. Create an rpc('get_mentor_activity_stats', {mentor_id: uuid}) that returns a single JSON object with all count fields. The Dart service then maps this JSON to PeerMentorStats. This is both more performant and more maintainable than orchestrating 3–4 separate Supabase queries in Dart.

For the watchStats stream, use a StreamController.broadcast() and call add() in invalidateCache() after clearing the cache entry — consumers can then call getStats() to get fresh data. The PeerMentorStats model should be immutable (const constructor with copyWith) so that badge-evaluation-service can safely store a snapshot without risk of mutation.

Unit tests with mocktail mocking all repository dependencies: (1) empty mentor returns PeerMentorStats.empty(); (2) correct aggregation with mixed activity types; (3) unique contact count de-duplicates repeated contact IDs; (4) cache hit on second getStats call — repositories called exactly once; (5) invalidateCache followed by getStats triggers fresh repository calls; (6) watchStats emits after invalidation. Integration test against Supabase test instance: seed known activity data and assert aggregation results match expected counts. Minimum 90% branch coverage.