fix: respect LLMRanker top_k on fallback

[charan-rathore]

Related Issues

• Fixes LLMRanker fallback paths ignore top_k and return all input documents #13045

Proposed Changes:

LLMRanker limits successful results to top_k, but returned every input document after generation or parsing failures and invalid queries. Apply the same limit to those fallback results in both run and run_async.

Document order and deduplication stay unchanged. The runtime top_k override still takes precedence. Documentation and a release note explain that fallback output now follows the same limit as successful ranking.

How did you test it?

• New regression cases: 16 failed and 8 passed before the fix.
• LLMRanker and LostInTheMiddleRanker suites: 99 passed, 3 live-provider tests skipped.
• Focused mypy: no issues in the implementation and test files.
• All applicable pre-commit hooks passed on the five changed files.
• A real Haystack Pipeline returned one fallback document with top_k=1, synchronously and asynchronously, without an LLM call.

Notes for the reviewer

The seven existing fallback tests with top_k=1 expected both input documents. Those expectations now follow the documented limit. New cases cover generation and parsing failures, empty and non-string queries, duplicate inputs, constructor defaults, runtime overrides, and limits larger than the input list.

This is an intentional result-limit change for callers relying on all documents during fallback. Setting top_k to the input document count keeps all fallback documents.

Instinct assisted. This patch was generated with assistance; human review and understanding are still required before submission.

reno lint could not finish in the shallow clone because history object f4e004c34cbc724b3ffb026debdf7a0541e4a21f is missing. Release-note YAML and backtick hooks passed. Full repository tests and type checks were not run.

Checklist

• I have read the contributors guidelines and code of conduct.
• I have updated the related issue with new insights and changes. No public comments have been made.
• I have added unit tests and updated the docstrings.
• I've used a conventional commit type for the PR title.
• I have documented my code.
• I have added a release note file.
• I have run pre-commit hooks and fixed any issue on changed files.

[vercel]

@charan-rathore is attempting to deploy a commit to the deepset Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

All committers have signed the CLA.