fix(core): handle NotRequired/Required in TypedDict OpenAI function schema conversion

[Alvin Tang (alvinttang)]

Description

Fixes #34085.

When a TypedDict with NotRequired[X] or Required[X] fields is passed to
convert_to_openai_function() (or .with_structured_output()), a TypeError
is raised inside _convert_any_typed_dicts_to_pydantic:

TypeError: NotRequired accepts only a single type. Got (<class 'str'>,).

Root cause

_convert_any_typed_dicts_to_pydantic has two problems:

1. Generic type handler (line 278): when it encounters NotRequired[str] as
   a generic type it rebuilds it as NotRequired[(str,)] (tuple), but
   NotRequired / Required are single-argument special forms and reject tuples.

2. TypedDict field loop: get_type_hints(include_extras=True) preserves
   NotRequired[X] wrappers on fields. The loop never stripped this wrapper, so
   the raw NotRequired[str] fell through to the generic handler and triggered the
   bug above. Additionally, NotRequired fields were incorrectly marked as
   required (default=...) in the Pydantic v1 model.

Fix

In langchain_core/utils/function_calling.py:

1. Field loop — before processing each annotation, unwrap NotRequired[X]/
   Required[X] and use default=None for NotRequired fields (and total=False
   fields without an explicit Required wrapper) so they are absent from the
   generated JSON schema required array.

2. Generic handler — when origin is NotRequired or Required, index with
   the single unwrapped arg (type_args[0]) rather than the full tuple.

Testing

Three new parameterised test cases (each run against both typing.TypedDict
and typing_extensions.TypedDict):

• test_convert_typed_dict_notrequired_fields — exact repro from the issue
• test_convert_typed_dict_required_fields_total_false — Required[] on a
  total=False TypedDict
• test_convert_typed_dict_notrequired_nested_type — NotRequired[list[str]]
  (nested generic, the direct trigger of the original TypeError)

All 36 existing tests in test_function_calling.py continue to pass.

Checklist

• Tests added
• No regressions (full test_function_calling.py suite passes)
• Follows existing code style

[github-actions]

This PR has been automatically closed because you are not assigned to the linked issue.

External contributors must be assigned to an issue before opening a PR for it. Please:

1. Comment on the linked issue to request assignment from a maintainer
2. Once assigned, edit your PR description and the PR will be reopened automatically

Maintainers: reopen this PR or remove the missing-issue-link label to bypass this check.