by Ken Ashcraft's report.  I think there is no actual bug here since if
the int32 value does wrap a little bit, palloc will still reject it.
Still it's better that the code be obviously correct.