Different topic assignment on training data when using saved model

Nobody has claimed this yet.

### Have you searched existing issues? 🔎

- [X] I have searched and found no existing issues

### Desribe the bug

When I save a model with pytorch serialization, then use the model to transform the training data, the new topic assignment is different from the "old" topic assignment.

### Reproduction

```python
from bertopic import BERTopic

topic_model_new = BERTopic.load("model")
# old topic assigment
new_df = topic_model_new.get_document_info(abstracts)
# new topic assignment
topics, probs = topic_model_new.transform(abstracts, embeddings)
(new_df['Topic'] == np.array(topics)).value_counts()
```

> Topic
> True 1168
> False 157
> Name: count, dtype: int64

### BERTopic Version

0.16.3

Start with the BERTopic.load, get_document_info, and transform calls shown in the reproduction, using the saved model and training data described there. Compare the topic assignments returned by both paths and trace where they diverge; done means the saved model produces consistent assignments for the same training data.

Written by the indexing model from the issue text.