Follow-up to v1.44.0's self-update queueing, answering @famewolf's question "what happens if container updates fail with a queued selfupdate afterward?" (#2):
- Per-container failures (rollback, "left in place") are normal batch results — reported first, and the queued self-update still runs afterwards.
- A flow-level crash (exception in the batch machinery) now cancels the queued self-update with an honest message instead of restarting on top of an unreported error.