Keep hook_cron to id selection: queue settlement re-submissions, expiry cancellations and suspensions

Problem/Motivation

Three parts of hook_cron do unbounded synchronous remote work before core gets to run the queues:

• CycleEngine::settleOpenCharges() loads every stale charge and re-submits each to its connector inline.
• The expiry sweep calls CycleEngine::end() per expired charge-based subscription, which calls the remote cancellation service, one connector call per row.
• DunningManager::sweepSuspensions() loads all past-due subscriptions whose grace has ended and, per row, saves the user, dispatches an event and sends mail.

All three use a single loadMultiple() with no range. Under automated cron (web-request time limit) or a slow provider, cron is aborted before processQueues(), so billing, retries and login reconciliation stall precisely during the provider outage the sweep exists for.

Proposed resolution

• Cron selects ids and enqueues: a settlement queue (or the cycle queue with an item type) for stale charges, the cycle queue for expiries, and a dunning queue item for each suspension. The existing per-item isolation and the per-subscription lock from #3630967: Lock the cycle per subscription, stop re-enqueuing past-due subscriptions, and refuse a pending charge result without a remote id apply.
• Test: a throwing connector in the settlement path does not prevent the cycle queue from being drained in the same cron run; each sweep's existing Kernel tests keep passing through the queue.

Remaining tasks

• Design pass on queue layout; implementation; tests; no change record unless Drush options change.

User interface changes

None.

API changes

Possibly new queue worker plugin ids.

Data model changes

None.