New write batches currently start from where the last one completed.
We have no idea where the head is after switching batches, so this
makes little sense. Instead, start the next batch from the request
with the earliest deadline in the hope that we avoid a deadline
expiry later on.