test: enable mypy typing checks for test/components/samplers/

[sainikhiljuluri]

Related Issues

Relates to #10396

Proposed Changes

Adds test/components/samplers/ to the types mypy allowlist, continuing the effort in #10396 to type-check our test files. The directory contains only __init__.py and test_top_p.py, so this covers it completely.

Both baseline errors were the same one:

test/components/samplers/test_top_p.py:68: error: Value of type variable "SupportsRichComparisonT" of "sorted" cannot be "float | None"  [type-var]
test/components/samplers/test_top_p.py:87: error: Value of type variable "SupportsRichComparisonT" of "sorted" cannot be "float | None"  [type-var]
Found 2 errors in 1 file (checked 2 source files)

Document.score is float | None, which isn't orderable, so sorted([doc.score for doc in docs]) is rejected. I narrowed with if doc.score is not None — the idiom already used for this exact expression in test_multi_query_text_retriever.py:149,191, test_multi_query_embedding_retriever.py:279 and test_multi_query_text_retriever_async.py:70,178.

Why one site gets an extra assertion and the other doesn't

The filter on its own would have quietly weakened test_run. The unfiltered sorted() was incidentally enforcing "every document has a score" by raising TypeError, and filtering throws that away. Concretely, if a document lost its score, three scores would filter to two, and sorted_scores[:2] would still match the two filtered documents — every assertion in the test would pass. So test_run also asserts the narrowed list kept every document, which restores exactly the invariant the TypeError used to provide.

test_run_top_p_1 needs no such guard: the pre-existing assert len(docs_filtered) == len(docs) above it already ties both sides to the input length, so a dropped score makes the lists different lengths and fails the comparison.

I also considered # type: ignore[type-var]. It's shorter, but AGENTS.md asks to avoid type: ignore where possible and warn_unused_ignores = true makes ignores a maintenance liability. assert all(s is not None for s in scores) doesn't work — it doesn't narrow the comprehension's element type, and mypy still errors.

How did you test it?

hatch run test:types                      # Success: no issues found in 2 source files
hatch run test:unit test/components/samplers/   # 14 passed

• Behaviour is unchanged: all 14 tests pass both with and without this change. I verified the revert actually applied before trusting the result.
• Ran the full types scope (449 source files) to confirm nothing else regressed.
• ruff format --check and ruff check are clean.

No release note: this is a tests-only change, matching the precedent in #12327, #12394 and #12333.

Checklist

• I have read the contributors guidelines and the code of conduct
• I have updated the related issue with new insights and changes
• I've used one of the conventional commit types for my PR title
• I documented my code
• I ran pre-commit hooks and fixed any issue

---

Note

This contribution was AI-assisted (Claude Code). The analysis, the mutation testing behind the assertion argument above, and the verification commands were run and reviewed before submitting.

[vercel]

@sainikhiljuluri is attempting to deploy a commit to the deepset Team on Vercel.

A member of the Team first needs to authorize it.