Citator Experiments: Fine-grained classification

[rachlllg]

All prior experiments will be conducted on (simpler) binary classification (overruled or not overruled), once Eyecite enhancements are in place, hopefully the excerpts will include all the details for the model to make more fine-grained classifications (followed, distinguished, partially overruled, fully overruled, rejected, mentioned, etc).

This is a blanket ticket for experiments needed to achieve this. Ideally, we will use the same model as we selected for binary classification but enhance it by tuning the prompt and/or finetuning the model to achieve this.