Fix TypeError in save() when TopicMapper.mappings_ contains None (#2432)

[lntutor]

What does this PR do?

Fixes #2432

Root cause

TopicMapper.add_new_topics pads the intermediate columns of a newly added row with None:

to_append = [key] + ([None] * (length - 2)) + [value]

As long as the mapping matrix has only two columns this is harmless (length - 2 == 0, so no None is inserted). However, once the matrix grows past two columns — which happens after repeated partial_fit calls — every new topic row gets one or more None placeholders. mappings_ is then no longer a homogeneous integer matrix.

Saving the model later calls _save_utils.save_topics, which converts the whole table with:

np.array(model.topic_mapper_.mappings_, dtype=int)

and this raises TypeError: int() argument must be ... not 'NoneType', exactly as reported in the issue's partial_fit + save(serialization="safetensors") reproduction.

Fix

A brand-new topic maps from itself in all prior states, so the intermediate columns should hold the topic id rather than None:

to_append = [key] * (length - 1) + [value]

This keeps mappings_ a homogeneous integer matrix so save() succeeds, and it is a no-op for the common two-column case ([key] * 1 + [value] == [key, value], identical to the previous output). On load, mappings_ is assigned back verbatim, so the rows round-trip unchanged.

Test

Adds test_topic_mapper_add_new_topics_keeps_integer_matrix in tests/test_bertopic.py. It builds a TopicMapper, widens the matrix past two columns to mimic repeated partial_fit states, calls add_new_topics, and asserts that np.array(mappings_, dtype=int) no longer raises, that pre-existing rows are unchanged, and that the new rows are homogeneous integers. The test is lightweight and does not train a model. It fails on the previous implementation (with the TypeError from the issue) and passes with this change.

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? See model.save() fails with TypeError when topic_mapper_.mappings_ contains None #2432.
• Did you make sure to update the documentation with your changes (if applicable)? Not applicable.
• Did you write any new necessary tests?