llama-swap emits inflight events with operation 'upsert' (a single request,
re-emitted as elapsed_ms updates) and id-only 'remove' events. The decoder
ignored 'upsert' and dropped every id-only remove, so the tracker only ever
reflected the connect-time snapshot and the count got stuck (e.g. at '1').
The tracker now keys in-flight state by request id (snapshot/add replace by
id; remove deletes by id), and the decoder normalizes 'upsert' to 'add' and
extracts the id from bare removes.
The key now renders the live request count (large) when a model is active,
plus a 1Hz count-trend sparkline of the last 60s in a bottom strip. IDLE/
LOADING/OFF/OFFLINE states unchanged. History lives in per-key state with
a sampler cleared on disappear.
bzuccaro