I’m trying to understand how OpenAI’s enforced hard spending limits behave when thousands of API requests are submitted concurrently. The documentation mentions that spending can slightly exceed the configured limit while enforcement propagates. How are…