Learn how multi-vector search stores multiple vectors at a single embedding path and uses MaxSim to retrieve documents with token-level or passage-level relevance.