fix: preserve Field constraints on variadic tool parameters

[rioyu123]

This pull request fixes lost Pydantic Field constraints on variadic tool parameters.

For example, *scores: Annotated[int, Field(ge=0)] currently produces an unconstrained item schema and allows negative values to reach the tool callable. The same issue affects values collected through **kwargs.

The fix preserves the existing Field constraint metadata on each positional item or keyword value when building the Pydantic model. As a result, the generated JSON schema and runtime validation both enforce the declared constraints, and invalid values are rejected before the callable runs.

This is intentionally narrower than #4609: it only covers the existing supported Annotated[T, Field(...)] form for *args and **kwargs. It does not add support for arbitrary outer Annotated metadata.

Ordinary parameters, defaults and aliases, argument reconstruction, description precedence, existing tuple handling, and strict-schema behavior for arbitrary-key **kwargs remain unchanged.

Test plan

• Added regression coverage for *args and **kwargs constraints in both generated schemas and actual tool invocation.
• Covered numeric bounds, string length constraints, valid and invalid values, omitted/empty variadic arguments, and existing tuple behavior.
• Full repository verification passes, including formatting, linting, Mypy, Pyright, and the test suite.
• Python 3.10 smoke checks pass for both variadic forms.

Related context

#4609

Checks

• Added regression tests.
• Ran .agents/skills/code-change-verification/scripts/run.sh.
• Confirmed the full verification suite passes.