[yonigozlan]

When tool schemas are passed to the response parser, an integer argument written as a whole-number float like "3.0" is now cast to 3 instead of being left as the raw string.

• JSON Schema (draft 6 onward, including 2020-12) defines integer as any number with a zero fractional part, so 3.0 is a valid integer value. jsonschema (Draft 7 and 2020-12 validators) and Pydantic v2 both accept it, and Pydantic returns 3.
• Models do sometimes write integer arguments this way, and the tool then receives a string where its schema declares an integer.
• Values with a fractional part ("3.5"), inf and NaN still fail the integer cast, so they fall through to the next declared type or stay raw, as before.
• Adds a unit test next to the existing argument coercion tests.
• Same change as in Transformers: Accept whole-number floats for integer tool arguments in parse_response huggingface/transformers#49547.

---

CI States

Latest PR Test (Base): ✅ Run #39535859047
Latest PR Test (Extra): ❌ Run #39535858929
Latest PR Test (AMD ROCm 10): ❌ Run #39535858894

[JustinTong0323]

LGTM