Flaky test: TestExtractLLMDoesNotBookASingleFlightFollowersSaving fails ~3 in 4 runs on main under load #479

Seen while reviewing #474: on origin/main the test failed 29 of 40 runs; on the #474 branch 31 of 40, with no change to the code it covers. It looks timing-dependent, most likely on a single-flight follower racing the leader under CPU load. It blocks a clean go test ./... on busy hosts. Next step: make the test wait on the single-flight result explicitly instead of on timing.