Confirm no signup users stranded by the Aug-9 lost auth emails (78 auth emails DLQ'd)

[mdenner1234]

Context

From the Aug-2026 activity-log audit. On 2026-08-09 ~01:29 UTC, 332 emails were dead-lettered ("Lost before send — reconciler timeout") when the restored stuck-email reconciler drained a pre-cutover backlog. Of those, 78 were auth emails: 70 signup confirmations, 7 password-recovery, 1 magic-link.

Root cause is already fixed — the reconciler cron (reconcile-stuck-emails) was missing since the Lovable→owned-project cutover; migration 20260809120010 restored it. It now runs every 5 min (2,217 successful runs at audit time) and there have been zero email_dlq events since Aug 9. This is not recurring.

Recovery/magic-link are self-service (the user just retries), so the only lasting risk is a signup user who never confirmed and therefore still can't log in.

Task

Confirm nobody is stranded, and re-trigger confirmation for any who are.

1. Run the read-only check (worktree activity-log-audit-aug):

   $env:PGPASSWORD="<prod db password>"; node _audit_stranded_signups.mjs
   

   • Sections 1–2 re-prove the reconciler is active and DLQ stopped after Aug 9.
   • Sections 3–3b count pre-Aug-9 accounts still email_confirmed_at IS NULL (potentially stranded).
   • Section 4 joins the Aug-9 DLQ signup recipients to auth.users for the direct-victim list.
2. For any confirmed-stranded signup users, re-send the signup confirmation (self-service resend or admin trigger).

Not needed

No code change for prevention — the reconciler + syshealth watchdog already cover it. This is a one-time data/outreach confirmation.

Ref: audit PRs #251 (captcha lockout), #252 (edge-deploy-smoke hardening).