fix: give each FileContent its own extra dict in FileToFileContent

[Vedant-Agarwal]

Related Issues

• N/A — small self-contained bug fix found while reading the converters.

Proposed Changes:

FileToFileContent.run() normalizes extra via normalize_metadata(), which returns [meta] * sources_count for a single dict and [{}] * sources_count for None — i.e. the same dict object repeated. The component then passed that object straight into every FileContent(..., extra=extra_dict), so all FileContents produced by one run() share one extra dict. Mutating one file's extra downstream (e.g. files[0].extra["page"] = 1) silently mutates every sibling's extra.

FileToFileContent is the only converter affected: the text/pdf/docx/tika/csv/markdown converters consume normalize_metadata() but merge into a fresh {**bytestream.meta, **metadata} dict per source, so the shared return is harmless for them.

The fix copies the dict per source (extra=dict(extra_dict)), matching the other converters' behaviour.

How did you test it?

• Added test_run_with_extra_dict_does_not_share_reference, which asserts the two FileContents don't alias the same extra and that mutating one doesn't leak into the other. It fails on main ({'tenant': 'acme'} is not {'tenant': 'acme'} → AssertionError) and passes with the fix.
• python -m pytest test/components/converters/test_file_to_file_content.py → all green.
• ruff check / ruff format --check clean on both files.

Notes for the reviewer

Minimal one-line change at the sink. I considered fixing normalize_metadata() to return independent dicts, but scoped it to the only broken consumer to avoid changing a shared utility that every other converter already handles safely.

This PR was written with the help of an AI assistant. I have reviewed the changes and run the relevant tests locally.

Checklist

• I have read the contributors guidelines and the code of conduct.
• I've used a conventional commit type for the PR title (fix:).
• I have added unit tests and documented my code.
• I have added a release note file.

[vercel]

@Vedant-Agarwal 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.