The Power of Distance Distributions: Cost Models and Scheduling Policies for Quality-Controlled Similarity Queries