fix(core): avoid noisy pydantic serialization warnings for model_dump items

[OiPunk]

Summary

• suppress pydantic serializer warnings when coercing model objects to API input dicts
• add a safe fallback path for model_dump-compatible objects that do not accept the warnings argument
• add regression coverage for container_file_citation-style payloads to ensure warning-free normalization

Problem

ensure_input_item_format / input item normalization can emit noisy PydanticSerializationUnexpectedValue warnings for mixed/legacy annotation payloads (for example container_file_citation variants). This muddies logs and affects common final_output/history handling workflows.

Fix

• introduced _model_dump_without_warnings(...) in src/agents/run_internal/items.py
• switched coercion paths to use this helper in both _coerce_to_dict and fingerprint_input_item
• helper first tries model_dump(exclude_unset=True, warnings=False) and falls back to model_dump(exclude_unset=True) for compatibility

Tests

• uv run ruff check src/agents/run_internal/items.py tests/test_agent_runner.py
• uv run mypy . --exclude site
• uv run pytest tests/test_agent_runner.py -k "ensure_api_input_item" -q
• uv run coverage run -m pytest tests/test_agent_runner.py -k "ensure_api_input_item" -q && uv run coverage report -m src/agents/run_internal/items.py

Fixes #772

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 54e0240704

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".