Clustering: Trim in-memory log on persistence recovery

At runtime, as state is applied and replicated, we trim the in-memory log to lower the memory footprint. However the persistent log is only trimmed when we do a real snapshot so, on restart, the recovered log will have a larger footprint. It will be trimmed as new state is subsequently applied but we should trim the in-memory log similarly as we're recovering the log.

We'll need to persist the replicatedToAllIndex. We should be able include it with the ApplyJournalEntries message.