The slow part is for CF-related constraints like long-running value
parsers. Also, the "fast" constraints and other problems are reported
first via a new checkpoint call to the problems collector. The complete
list of problems are then reported afterwards at the end of the
collection, like before.