use a k-means based cluster approach to speed up similarity searches

initial checkin. working vector storage and similarity search