markBatchFailureNoAttemptBump

Record a batch-level failure without bumping per-event attempts. Used when the whole batch fails for reasons outside the individual event (network transient, auth error, schema-level permanent). Without this, a single 30-minute outage processed as 50-event batches would march every event to maxAttemptsPerEvent and quarantine the fleet's backlog despite nothing being wrong with the events.

Still sets state=FAILED and records the error + next-attempt gate so the event is eligible again after backoff.