Send model session alerts to on-call ML engineers before standup

Your model sessions with prompt traces arrive unreviewed, causing missed quality regressions and delayed bug triage. Alerts let on-call ML engineers investigate flagged sessions so issues are triaged before morning standup.