Compare mode: Agreement measures

[melandresen]

For our current annotations, some additional agreement measures would be helpful. Right now, agreement is often 0, but it is usually not as bad as it looks at first sight. I am not sure about the perfect setup, but I would like to see a more generous score for cases

1. where spans are not identical, but overlapping, and
2. where spans are identical, but the labels differ
   (or both).

(This would be especially nice in combination with #362 )