Bot.listen awaited every frame to completion, so one Commons fetch or one
attachment upload held up every text reply behind it (DOCS-latency.md).
Each frame now runs in its own task, never more than LIMIT at once; the
reader waits for a slot before spawning, so a burst applies back-pressure
at the websocket instead of fanning out upstream. One person's commands
stay serialised (user_lock.py): two quick !kot would otherwise both pass
the cooldown gate before either recorded a use.
Every served command reports its timings split into the bot's own time
and everybody else's (served.py); /_selftest carries that block and
tool/latency.py holds own_ms to 100 ms at p95, with the command's
cooldown switched off for the run and put back afterwards.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
