Otherwise recent additional checks in tal() complain that we're freeing a take() pointer. In this case, we're exiting so it's harmless, but it's still a latent bug. Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>