Put unsynced usage back on the billing queue

Re-queues { usage_event_id } messages for ledger rows Stripe has not acknowledged, and RESETS their re-drive counter — the reason an operator presses this is that they have fixed the cause, and a counter that survived the fix would refuse the first retry after it. Idempotent and safe to repeat: the consumer derives the amount, customer and workspace from the persisted row and collapses a row that has since synced to a no-op. Nothing here can change an amount. Every call writes an audit row naming who did it.