Reducing the maintenance overhead of test_oat_hooks

I've been annoyed several times lately by having to update
the list of node types embodied in test_oat_hooks.c's
nodetag_to_string(). I got around to looking at that more
closely, and realized that it is only used for utility
statements, which (a) are a very small subset of the node
types that that function knows about, and (b) we already
have a mechanism to get string identifiers for, and (c)
those identifiers are already standard parts of our user API,
unlike the strings exposed by nodetag_to_string(). I do not
think that test_oat_hooks.c has any business imposing
an extra maintenance burden on us all, so I propose
nuking nodetag_to_string() from orbit, as attached.

(Incidentally, this improves test_oat_hooks's own
reported code coverage from 14.0% to 76.1%, because
so much of that switch is dead code.)

regards, tom lane