Add "labels" to SetFitModel, store/load from configuration file

[tomaarsen]

Hello!

Pull Request overview

• Add SetFitModel.labels, SetFitModel.label2id, SetFitModel.id2label.
• By default, use labels in model.predict. Can be disabled with model.predict(use_labels=False).
• Save normalize_embeddings and labels in a config_setfit.json file.
• Move config_setfit_span.json contents (i.e. span_context) to config_setfit.json.
• Fix for wrong device with training with a differentiable head.

Details

The core of this PR is to allow:

model = SetFitModel.from_pretrained("tomaarsen/setfit-bge-small-v1.5-sst2-8-shot")
preds = model.predict([
    "It's a charming and often affecting journey.",
    "It's slow -- very, very slow.",
    "A sometimes tedious film.",
])
print(preds)
# ['positive', 'negative', 'negative']

i.e. get string outputs.

• Tom Aarsen

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.