fix(tools): unwrap FieldInfo default to avoid pydantic warning

[chaynabors]

The @tool decorator emitted a PydanticJsonSchemaWarning whenever a parameter used Field(default_factory=...) because _extract_annotated_metadata rebuilt the field with Field(default=param_default, ...), nesting the original FieldInfo as the default value. The fix detects when param_default is already a FieldInfo and rebuilds from its default, default_factory, and validation kwargs so constraints like ge, le, and pattern are preserved.

Fixes #1914.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[awsarron]

please add test coverage

[mkmeral]

/strands review

[github-actions]

Assessment: Comment (minor changes suggested)

The fix is correct and well-targeted. I verified that main nests the FieldInfo as a default (producing the PydanticJsonSchemaWarning), while this branch reuses/merges it cleanly — ge/le/pattern constraints survive in both the generated JSON schema and at runtime validation, and the docstring Args description is still honored. The regression test correctly fails on main and passes here. Small, focused change that does one thing.

Review themes

• Test coverage: The new test exercises only the no-description merge branch; the description-already-present branch is uncovered (Codecov's missing lines). The test also asserts a single field rather than the full schema shape, and never asserts on items — the very default_factory parameter from [BUG] @tool emits PydanticJsonSchemaWarning for Field(default_factory=list) #1914.
• API consistency: param = Field(ge=0) now works, but Annotated[T, Field(ge=0)] still raises NotImplementedError with a note claiming constraints can't be reliably preserved — a premise this PR contradicts. Worth a decision on unifying the two paths or updating the note.

Nice, minimal fix that cleanly resolves the warning without changing the happy path.

[mkmeral]

/strands review