Identity sequences shouldn't be addressed directly by name in normal
use. Therefore, requiring them to be added directly to publications is
a faulty interface. I think they should be considered included in a
publication automatically when their owning table is. See attached
patch for a sketch. (It doesn't actually work quite yet, but it shows
the idea, I think.)
If we end up keeping the logical replication of sequences feature, I
think something like this should be added, too.