fix: preserve explicit runtime retriever overrides

[vinkiYu]

Summary

• Preserve an explicitly provided empty ilters={} in FilterRetriever, including synchronous, asynchronous, and Pipeline execution.
• Validate runtime window_size values in SentenceWindowRetriever consistently across constructor, sync, and async paths.
• Add regression tests and release notes.

Tests

• python -m py_compile passed for all changed Python files.
• Hatch/pytest were not run in the available environment: Hatch is not installed and the available Python is 3.8.6, below Haystack's supported range.

[vercel]

@vinkiYu 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.

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
haystack/components/retrievers
filter_retriever.py
sentence_window_retriever.py
Project Total

This report was generated by python-coverage-comment-action

[julian-risch]

@vinkiYu Thank you for opening this PR! I understand that there should be a simple way to reset filters in the run method to overwrite whatever filters were set in init. However, {} is not a valid filter dict in our filter spec because it has no operator/conditions key. Currently, a pipeline that builds its filter dict conditionally (ConditionalRouter, a Jinja template, an LLM-produced dict) and yields {} for the no-filter case falls back to the init filters. After this PR the same pipeline would return the entire document store, which is a breaking change with big impact.

As a workaround, you can use {"field": "id", "operator": "!=", "value": "__nope__"}, which works for all our different document stores. An alternative is to pass filters at runtime only and build the input dict conditionally. That way, the filters key is only set when needed:

inputs = {"filters": user_filters} if user_filters else {}
pipeline.run(data={"retriever": inputs})

We could also think about implementing a Haystack NO_FILTERS object. However, if we introduce it, we should ensure that it works with all retrievers in haystack-core-integrations too.

The existing workaround and the breaking change are the main reasons for me to close this PR.

---

Regarding the other change (window_size=0):
Could you please open a separate pull request and:

• Update run_async docstring according to your changes and mention window_size=0 is rejected
• Use hatch run release-note to generate release notes (see also AGENTS.md).
• Clean up the dead run(...) calls in the renamed init tests after the assert

Thank you so much! Looking forward to your PR!