Currently, using a failpoint in a test requires using the multi-threaded tokio runtime (tokio::test(flavor = "multi_thread")). If that's not specified, the test panics with an unclear error message; ideally we could query for the presence of the runtime and provide a friendlier error message.