Implement corpus-based evaluation

[johann-petrak]

Strategy:

1. export / extract the codes for the targets from set A
2. classify, then export/extract codes for the predictions from set B
3. use standard chunk evaluation code (huggingface?) to calculate metrics