Live-check the org's cached Stripe/Metronome ids against the vendor's real state and null whichever are confirmed gone, instead of trusting a cache that never learns about a change made by hand in Stripe or Metronome (e.g. a superadmin deleting a test Stripe customer in the Stripe dashboard). Used by the superadmin panel's Refresh button and by the delete-organization guard, so neither ever needs a manual SQL clear for an id the vendor has already dropped.

A Stripe customer is "gone" when the retrieve 404s or answers with deleted: true (Stripe keeps deleted customers retrievable forever with that flag, per the API — it does not 404 them). A Metronome contract is "gone" when the retrieve 404s, or "ended" when its own ending_before (the same field cache.py reads for term_end) is in the past — either is enough to clear both Metronome ids, matching the manual instructions this replaces (confirm the contract has ended; Metronome customers are not archived here).

Any other vendor error (5xx, timeout, auth) leaves the id untouched — only a confirmed answer clears anything.