That's the answer I was seeking. So it's case closed for both of my « Two questions about "pg_constraint" ». Thanks, Tom. And thanks to the others who've contributed to this thread.
My other question was about the "connamespace" column. It seemed to me, both at first and still now, that this is a clear instance of a transitive dependency.
I think a more productive way to think about it is that it's denormalization for efficiency; specifically to let constraints be looked up by name+namespace without having to get other catalogs involved. (SET CONSTRAINTS is one thing that requires that, and I think there are others.)