Function run_queue_dispatcher 

pub async fn run_queue_dispatcher(
    job_rx: Receiver<GenerationJob>,
    state: AppState,
)

Runs the multi-GPU dispatch loop. Routes each generation job to the best GPU worker per select_worker’s tier order: model loaded + idle > idle empty GPU that fits > model loaded but busy > idle empty GPU that does not fit > most-headroom fallback (evict LRU there). Uses a small lookahead buffer so an interleaved queue ([A, B, A, B]) doesn’t force a sibling worker to swap models when one already has the right one warm.

Exits when the sender half of the channel is dropped (server shutdown).