event_accepted volume drop diagnostic recipe + reprocess flag scope

Debugging recipe when a downstream service consuming event_accepted sees volume drop: (1) reprocess-only-events-not-in-nodestore — per-process CLI arg on dedicated reprocess consumers (events-reprocessing); mutually exclusive with reprocess-only-stuck-events. On main events.ingest-consumer impact is modest because dedup cache (ev:{project_id}:{event_id}, ~1h TTL) at processors.py:117-131 catches most repeats. (2) inline_save_event=True (5ae4736ff76, Jun 15 2026) — taskbroker passthrough; persistent failures → DLQ → signal never fires. (3) Organization.DoesNotExist at processors.py:246 — silent drop on deleted orgs. (4) UDP/transport loss — separate from signal firing; instrument event_accepted.send_robust() calls and compare against receiver arrivals. Gotcha: assumes drops correlate with flag; on main consumer the flag is a SECOND filter behind dedup, not the primary. Reference impl: getsentry orbital.py UDP receiver — verify by killing flag on one pod and watching for spike. Verify: ps aux | grep events | grep reprocess-only-events-not-in-nodestore.