feat: support pre-computed UMAP embeddings via umap_embeddings parameter [stacks on #2515]

[pidefrem]

What does this PR do?

feat: support pre-computed UMAP embeddings via umap_embeddings parameter

BERTopic always computes UMAP internally. While BaseDimensionalityReduction lets users skip UMAP entirely, there's no way to reuse specific UMAP embeddings from a previous run — which is needed for hyperparameter grid search (same UMAP, different HDBSCAN settings) and cross-session caching.

Changes:

Add a umap_embeddings parameter to fit(), fit_transform(), and transform(). When provided, skip the UMAP computation and use the given embeddings directly. The UMAP model is still stored (for .transform() on new data) but not called during that fit.

All new parameters default to None — existing code unchanged.

This is complementary to BaseDimensionalityReduction, not a replacement: it reuses specific UMAP embeddings without replacing the UMAP model.

Fixes #2504

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 #2515. This branch includes the representative-docs caching commits from #2515 as a prerequisite (marked [prereq]), so the diff here shows that work too. Once #2515 merges, I'll rebase onto master and the prereq commits drop out, leaving only the umap_embeddings change. Marking ready for review after #2515 lands.