Building a <50MB RAM background daemon in Go - how onWatch tracks 6 AI APIs concurrently