Fix extraction from multi-line Pretix PDF tickets

The old approach got the fields wrong in some of those cases. This
shouldn't happen anymore, at the expense of missing some details
(attendee name, ticket name) which cannot be uniquely identified from
the text.