Improve handling of long ref files, propose an alternative to raw logs
We've always had issues with ref files being too long or having too long lines. In such cases, the standard validators are taking too long and the refs are mostly useless as a human being cannot cope with the diff output when not empty.
So we need a way to handle and diff more cleverly bigger amount of data. One way is to be able to direct it to structured output (json ?) and have a more clever diff.