SGLang per-request time-per-output-token (TPOT) histogram

SGLang main, python/sglang/srt/observability/metrics_collector.py, read 2026-10-04. NOT ON BY DEFAULT: nothing is exported unless --enable-metrics (ServerArgs default False), which is also the only thing that mounts /metrics on the server port. Label keys are model_name, engine_type, tp_rank, pp_rank, moe_ep_rank (plus dp_rank under data parallelism, priority under priority scheduling, and anything in extra_metric_labels). Bucket ladders are overridable per histogram via get_histogram_conf_from_env. Source documentation states the definition verbatim: 'Per-request TPOT in seconds: (finished_time - first_token_time) / (completion_tokens - 1)'. Shared bucket ladder with the ITL histogram; the denominators differ, so the quantiles are not the same statistic.