The event log is the account/channel authority, so make it durable:
persist() now fsyncs every committed event before reporting success, and
a serialisation failure is an error instead of a silently-dropped blank
line. Add a hardened systemd unit (scripts/echo.service), an atomic
gzip backup script with retention (scripts/backup.sh), and a Deployment
wiki page (install, durability, backup/restore, upgrade, monitoring).
