This patch should help fix cases with two separate fk constraints
in a row that happen to reference the same pk constraint with
an on update cascade and similar cases. It now should detect
correctly that a pk row was added later after a delete or update
on a no action deferred fk and not incorrect error as well
as dropping a check on insert/update to the fk table if the
row we're being referred to no longer is valid (I'm using
HeapTupleSatisfiesItself because the comment implied it was
what I was looking for and it appears to work :) ).
I've got regression tests but I'm holding off on those until
we decide whether or not we're dropping the triggered data change
errors since a couple of the tests would hit that case and I'll
either change them or drop them if we're not dropping the error.