Preserve node predictions until server catches up

Preserve node predictions until server catches up

Apply implicit sequence numbers on actions sent by clients, and use them
for the server to indicate, in a node status update, how up to date it is
in processing the client's actions.  Rather than apply all node updates
blindly, the client now checks whether it has relevant predictions that
are more recent than what the server has used to produce the update,
and keep those predictions applied until the server has caught up.
This avoids lag-induced node glitches.