Two stabilization fixes after live testing showed ingest was invisible
and could amplify feed load:
1. Diagnostics were using devLogger.log, which is gated by IS_DEV_MODE
(false in all current builds) — so the [nip71Ingest] subscribe/flush
lines never printed and the feature was impossible to observe. Switched
the subscribe + per-flush lines to userLogger.info so we can confirm in
the console whether it subscribed (author count) and whether events are
arriving/injecting. (Temporary while stabilizing.)
2. Coalesce ingest-driven "videos:updated" signals behind an 8s throttle.
A burst of incoming events previously emitted a refresh per flush; each
refresh re-runs the feed and re-subscribes moderation/social context for
every video, which can contribute to a REQ storm. Injected videos are
already in the active store, so throttling only delays the visual
refresh, not the data.
Tests: throttle coalescing test (one immediate refresh per window, both
videos still stored); mutation-verified. 1234/1235 unit (known flake).
Co-Authored-By: Claude Opus 4.8 <[email protected]>
