Probabilities are NoneType when doing supervised learning

Nobody has claimed this yet.

I'm following the guide [here](https://maartengr.github.io/BERTopic/getting_started/supervised/supervised.html) on how to use BERTopic for supervised learning on the `20newsgroups` dataset. I want to see all calculated probabilities so I added `calculate_probabilities=True` to the BERTopic instance, see below. However, `probs` comes up as `NoneType`. Is some dependency I am lacking or something wrong with my implementation?

`probs` does return an array of probabilities when doing unsupervised learning, so I wonder if this issue is unique to supervised learning.

```
# Skip over dimensionality reduction, replace cluster model with classifier,
# and reduce frequent words while we are at it.
empty_dimensionality_model = BaseDimensionalityReduction()
clf = LogisticRegression()
ctfidf_model = ClassTfidfTransformer(reduce_frequent_words=True)

# Create a fully supervised BERTopic instance
topic_model= BERTopic(
umap_model=empty_dimensionality_model,
hdbscan_model=clf,
ctfidf_model=ctfidf_model,
calculate_probabilities = True
)

topics, probs = topic_model.fit_transform(docs, y=y)
```

Reproduce the supervised-learning example from the linked guide with the 20newsgroups dataset, LogisticRegression, and calculate_probabilities=True. Trace the supervised fit_transform path to determine why probs is None while unsupervised learning returns an array. Done means the behavior is fixed or the supervised limitation is clearly documented, with the reproduction covered by a regression test if the project provides one.

Written by the indexing model from the issue text.