Lemmatize when evaluating

[jdelaunay]

Lemmatize the predicted terms, and the unique annotations lists, for the evaluation. Test again the models:

KB:

• [x] XLMR
• [x] XLMR large
• [x] RoBERTa
• [x] RoBERTa large

Human:

• [x] XLMR
• [ ] XLMR large
• [x] RoBERTa
• [x] RoBERTa large