chore(examples): log network RTT to separate transport from pipeline latency

glass2glass much larger than rtt means the media pipeline (encode, jitter
buffer, geography of the extra cloud hop) is the cost. glass2glass close to
rtt means it is just the network round trip. This tells the two apart.

Author: pham-tuan-binh

examples/python/modal-mock-inference/robot.py

@@ -144,7 +144,7 @@ def on_action(action: Action) -> None:
                     f"replies={stats.replies} dropped={stats.dropped} "
                     f"glass2glass={fmt_us(stats.glass_us)} codec_lag={fmt_us(stats.codec_us)} "
                     f"e2e={fmt_us(m.policy.e2e_us_p50)}/{fmt_us(m.policy.e2e_us_p95)} "
-                    f"active={robot.active_operator()}"
+                    f"rtt={fmt_us(m.rtt.rtt_us_last)} active={robot.active_operator()}"
                 )
                 last_log = now
 
@@ -161,6 +161,7 @@ def on_action(action: Action) -> None:
         print()
         print("[robot] final:")
         print(f"  glass-to-glass p50/p95: {fmt_us(m.policy.e2e_us_p50)} / {fmt_us(m.policy.e2e_us_p95)}")
+        print(f"  network rtt (last):     {fmt_us(m.rtt.rtt_us_last)}")
         print(f"  codec lag (last):       {fmt_us(stats.codec_us)}")
         print(f"  frames sent:            {DURATION_S * FPS}")
         print(f"  replies / dropped:      {stats.replies} / {stats.dropped}")