CER-based OCR evaluation code

[keighrim]

We discussed evaluation schemes for the whole pipeline of

1. frame detection
2. text localization
3. document layout understanding
4. OCR , but couldn't figure out a good way for one-shot evaluation for such a long pipeline.

This issue is only restricted to the last component of the above pipeline and for the first round of evaluation, we will use https://github.com/clamsproject/aapb-annotations/blob/80a36781fd55b5b8cb74f2de187be026f1ecbb7f/newshour-chyron/golds/batch2/2022-jul-chyron.csv dataset as the evaluation data.

[keighrim]

closed via #26