fix: drop stale topic aspects when topics are recalculated

[raashish1601]

What does this PR do?

Fixes #2531

update_topics() sets representation_model to whatever is passed in, so after update_topics(docs, vectorizer_model=...) it is None. A later merge_topics() (or topic reduction) then recalculates the main representation but not the aspects. The old entries in topic_aspects_ stay as they were and keep pointing at the old topic ids, so after a merge every topic past the merged one shows the aspects of a different topic.

_extract_topics() now clears topic_aspects_ before recalculating representations for a new set of topics. Aspects that are still in representation_model get recalculated as before. Aspects whose model is gone are dropped instead of being left attached to the wrong topics.

Calling update_topics() by itself doesn't change the topics, so the aspects it keeps still line up and that path is unchanged.

Two tests in tests/test_representation/test_representations.py:

• merging topics recalculates the MMR aspect for the new topics (already worked, now covered)
• merging after update_topics() with a custom vectorizer leaves no aspect keyed by old topics. This one fails on master.

I ran tests/test_representation, tests/test_bertopic.py, tests/test_variations and tests/test_reduction on Linux (Python 3.12), plus ruff check and ruff format --check.

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. (Topic aspects are not updated when merging topics and the model uses a custom vectorizer #2531, labeled as a bug)
• Did you make sure to update the documentation with your changes (if applicable)? (not applicable)
• Did you write any new necessary tests?