feat: add FeatureImportance representation model with contrastive and embedding-aware methods

[pidefrem]

What does this PR do?

feat: add FeatureImportance representation model with contrastive and embedding-aware methods

BERTopic's topic term importance is based solely on c-TF-IDF. This PR adds a FeatureImportance representation class with two pluggable importance methods:

Method	Algorithm	Answers
"fighting_words"	Bayesian log-odds with Dirichlet priors (Monroe et al. 2008)	"What terms distinguish this topic from the rest?"
"centroid_distance"	Cosine similarity between topic centroid and term embeddings	"What terms are semantically central to this topic?"

Changes:

New FeatureImportance class under bertopic/representation/ that follows the existing BaseRepresentation interface. Slots into BERTopic's topic_aspects_ infrastructure:

from bertopic.representation import FeatureImportance

topic_model = BERTopic(
    representation_model={
        "Main": main_model,
        "FightingWords": FeatureImportance(method="fighting_words"),
        "CentroidDistance": FeatureImportance(method="centroid_distance"),
    }
)
topic_model.get_topic(0, aspect="FightingWords")

No changes to core BERTopic code — this is a new representation model that plugs into the existing pipeline.

Fixes #2506

Coordination note: this is purely additive — a new representation/ class behind the existing BaseRepresentation interface, no core changes. I shipped two methods (fighting_words, centroid_distance); if you'd prefer to start with just one to keep the surface minimal, or want a different class/method naming, happy to trim before review.

Before submitting

• This PR fixes a typo or improves the docs (if yes, ignore all other checks!).
• Did you read the contributor guideline?
• Was this discussed/approved via a Github issue? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes (if applicable)?
• Did you write any new necessary tests?