Force sync on exit regardless how long replication was running

[hifi]

Sync on exit isn't very robust on v0.3.9 or current main. If the replication is signaled immediately or the exec command terminates very quickly it's possible the database connection is never initialized and will never perform the final sync since it wasn't yet open. This is mainly possible because the connection is lazily initialized during the db Sync() call making the opening a bit racy and the close handler avoiding final sync if the database wasn't open.

This is easily reproducible by a command line like this:

litestream replicate -exec /bin/true

To make it succeed you can use a process that hangs around a little longer depending on replication target initialization speed:

litestream replicate -exec 'sleep 1'`

We're using this to force a one off sync where we'd like the Litestream exit code to be non-zero if there's any issues with the replication target so we can hold off any operations that depend on successful replication outside normal operation.