`model.save()` fails with `TypeError` when `topic_mapper_.mappings_` contains `None`

Have you searched existing issues? 🔎

• I have searched and found no existing issues

Describe the bug

When calling TopicMapper.add_new_topics, the method appends new rows to mappings_ with None placeholders:

def add_new_topics(self, mappings: Mapping[int, int]):
    length = len(self.mappings_[0])
    for key, value in mappings.items():
        to_append = [key] + ([None] * (length - 2)) + [value]
        self.mappings_.append(to_append)

This works during runtime, but later when saving the model with model.save() (serialization="safetensors"), it fails because _save_utils.save_topics tries to cast the entire mappings_ table to np.array(..., dtype=int):

File ".../bertopic/_save_utils.py", line 442, in save_topics
    "topic_mapper": np.array(model.topic_mapper_.mappings_, dtype=int).tolist(),
TypeError: int() argument must be a string, a bytes-like object or a real number, not 'NoneType'

Reproduction

from sklearn.datasets import fetch_20newsgroups
from sklearn.cluster import Birch
from bertopic.vectorizers import OnlineCountVectorizer
from bertopic import BERTopic

# Prepare documents
all_docs = fetch_20newsgroups(subset="train",  remove=('headers', 'footers', 'quotes'))["data"]
first_docs = all_docs[:50] # Making it small so that new clusters emerge with partial_fit and new mappings are added to the topic mapper
doc_chunks = [all_docs[50:][i:i+1000] for i in range(0, len(all_docs[50:]), 1000)]

# Prepare sub-models that support online learning
cluster_model = Birch(threshold=1.5, n_clusters=None)
vectorizer_model = OnlineCountVectorizer(stop_words="english", decay=.01),

# Train model for a first time
topic_model = BERTopic(
    language="multilingual",
    hdbscan_model=Birch(threshold=1.5, n_clusters=None),
    vectorizer_model = OnlineCountVectorizer(stop_words="english", decay=.01),
)

topic_model.fit_transform(documents=first_docs)

# Incremental fitting
for batch in doc_chunks:
    topic_model.partial_fit(batch)

topic_model.save("temp", serialization="safetensors") # -> This throws the error

BERTopic Version

0.17.3

Start with TopicMapper.add_new_topics and save_utils.save_topics, especially the np.array conversion shown in the traceback. Run the supplied incremental-fitting reproduction and inspect how model.save(..., serialization="safetensors") handles mappings containing None. Done means the model saves without the TypeError for this case.