fix(core): raise a clear UserError when a tool parameter is named model_config

[coderdailyone]

Summary

A tool parameter named model_config made function_schema() fail with an unrelated-looking error from inside Pydantic:

TypeError: 'FieldInfo' object is not iterable

function_schema() builds the argument model with create_model(f"{name}_args", __base__=BaseModel, **fields), and Pydantic interprets a model_config keyword to create_model() as the model's configuration dict rather than as a field, so it tries to iterate the FieldInfo. Other reserved names already produce readable Pydantic errors (model_dump / model_validate → "conflicts with member … of protected namespace", leading underscores → "Fields must not use names with leading underscores"); model_config was the one that failed opaquely.

This PR checks for that name before calling create_model() and raises a UserError that names the function and the parameter:

Parameter `model_config` in function configure is reserved by Pydantic and cannot be a tool argument. Rename the parameter.

No behavior changes for any other parameter name.

Test plan

• test_model_config_parameter_raises_a_clear_user_error in tests/test_function_schema.py: on main it fails with the Pydantic TypeError; with this change it gets the UserError.
• tests/test_function_schema.py, tests/test_function_tool.py, tests/test_function_tool_decorator.py: 177 passed. ruff format/ruff check clean, mypy and pyright clean on the changed files (Linux, Python 3.10). The repository-wide mypy src reports pre-existing Python 3.10 errors in archive_ops.py / run_loop.py unrelated to this change, so the full-stack checkbox is left unchecked.

Issue number

None (found while probing function_schema() with reserved parameter names).

Checks

• I've added new tests, if relevant
• I've run the verification steps from .agents/skills/code-change-verification individually (format, lint, typecheck on changed files, tests)
• I've confirmed all verification steps pass (see Test plan)
• If using Codex, I've run /review before submitting this PR

🤖 Generated with Claude Code

https://claude.ai/code/session_01DFSMrLZZ3oq1dKmJFAofz6

[sylvesterkaczmarek]

Catching model_config before create_model() is the correct ownership boundary: this is a Pydantic metaclass/config namespace collision, not a malformed tool schema, so allowing the opaque FieldInfo TypeError to escape misdiagnoses the user's function. The exact-name check is intentionally narrow and leaves all other parameter handling unchanged. The regression also pins the public error to the function and offending parameter.

[seratch]

We consider #4800 as the primary candidate for the same purpose.