perf: cache representative_docs_ to avoid redundant recomputation [stacks on #2496]

[pidefrem]

What does this PR do?

perf: cache representative_docs_ to avoid redundant recomputation

_extract_representative_docs() runs the full computation pipeline (sampling → c-TF-IDF → cosine similarity → MMR selection) every time it is invoked. In interactive workflows, multiple visualization calls each trigger this computation even though topic assignments haven't changed.

This is the root cause behind #2367 — KeyBERTInspired recomputes representative docs embeddings even when precomputed embeddings are available.

Changes:

Cache representative_docs_ after first computation. Invalidate the cache when topic assignments change (i.e., after update_topics(), reduce_outliers(), merge_topics()). Subsequent calls return the cached result without recomputation.

This follows the same pattern BERTopic already uses for topic_embeddings_ — compute once, store, invalidate on change.

Benchmark (to be filled in before submission):

Workflow (N repeated accesses)	Time before (no cache)	Time after (cached)	Speedup
20× representative-docs access, ~3k docs	(TODO)	(TODO)	(TODO)

First call costs the same; every subsequent access in an unchanged model is served from cache. The win is in interactive/dashboard workflows that re-trigger the computation repeatedly.

Fixes #2499

Recommended merge order: Best reviewed after #2496 (dedup fix) and #2496 (indexing fix) for _extract_representative_docs, so the cache stores correct results. The caching mechanism itself is independent.

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?

---

⚠️ Draft — stacks on #2496. This branch includes the _extract_representative_docs fix commits from #2496 as a prerequisite (marked [prereq]) so the cache stores correct results; the diff here shows that work too. Once #2496 merges, I'll rebase onto master and the prereq commits drop out, leaving only the caching change. Marking ready for review after #2496 lands.