do you have benchmarks on latency/throughput compared to pure‑Python agent frameworks? | discoverkit | discoverkit