Tokens diff view for contrastive attribution methods

[gsarti]

Description

This PR adds the original → contrastive label currently used in the PairAggregator to mark tokens that differ between the original generated sequence and the contrast_target when a contrastive attributed_fn is used in model.attribute. Previously, only the original target was shown. This enables to visually assess whether the alignment of contrastive target pair is correct.

Example:

import inseq
import pandas as pd

model = inseq.load_model("Helsinki-NLP/opus-mt-en-it", "saliency")
out = model.attribute(
    "UN peacekeepers",
    "I soldati della pace dell'ONU",
    attributed_fn="contrast_prob_diff",
    contrast_targets="Le forze di pace delle Nazioni Unite"
)
out.show()

• 💥 Breaking change: The TokenWithId items in targets will now have the new text to mark the contrastive step if they differ from the original when a contrastive step function is used. Their id is set to -1 to represent the fact they do not correspond to real tokens.