Optimize data aggregation flow (affects CPU + RAM + ingress)

Currently, all user data is loaded/cached in order to simplify the aggregation flow. We should implement a dependency analyzer that automatically calculates which rows are actually needed for the different aggregation algorithms.