The `conversations_count` we previously used counts replies, but also
all conversations that were tagged or re-tagged during the month – with
made counting the actual work spent on user support brittle.
Counting the replies is a better estimation of what we get.
Unfortunately this also removes the filtering-by-tag feature. To
mitigate this, the reports are now scoped to a specific mailbox. This
allows to create extra mailboxes for conversations that should't be
counted in the stats.
