Scan the entire merge requests table in batches of 100K and
then select only records that can contain mentions from each
batch. This way the number of fetched records is constantly
100K and we can ensure query does not timeout due to a overly
large batch of records being queried.