Embedding-based deduplication

Vector similarity to merge near duplicates and aliases