the persistence driver
the appliers, by row kind, that take another process's rows into this process's memory
when a finite number, the high-water to start from instead of the log's current head
a promise of { coordinating, from }, or { coordinating: false }
with an error when coordination is off or failed
Starts following one change log.
Called after the whole store has been restored. Registers the per-process pull job and the change-log trim job, and listens for the driver's nudge. A change log that cannot be read is not fatal: the process runs uncoordinated and keeps trying.