Investigate OEM adapter noise spike on 2026-07-30

Subtask of CIR-1190. Establish timeline and actual causal producer for OEM adapter queue/noise spike using Datadog, Slack incident evidence, and read-only production data. Separate confirmed facts from hypotheses. No production mutations. Progress 2026-07-30: scheduler input volume flat versus Jul 27/28 (Tesla ~234k/9h, texture ~101k/9h; Jul 29 Tesla 229,738, texture 100,406). No producer/import spike observed. First delayed-job burst coincides with OEM rollout at 19:30Z (sha-ef048fb5→sha-7f1e03e3), 16 Redis metrics-probe reconnects, and CrashLoopBackOff on OEM pods. Repeated delayed-due bursts followed pod churn; verifying exact trigger/recovery mechanics.