Backdoor Learning Curves: Explaining Backdoor Poisoning Beyond Influence Functions