Improve data model to cope with bigger logs (migrate to streaming logs)

[finfinack]

Currently, a log is read in full and then processed in memory. While this is fine as long as the logs stay small, this will have to move to a streaming approach at some point.