Fix rec_strict_json_schema to handle numeric constraint keywords

[yurekami]

Summary

• Fixed rec_strict_json_schema function to handle int and float values as terminal types
• This fixes issue [BUG CLIENT]: response_format_from_pydantic_model can't accept integer keyword #300 where Pydantic models with constraint keywords like min_length, max_length, minItems, maxItems, minimum, maximum, and multipleOf would cause a ValueError

Problem

When using response_format_from_pydantic_model() with Pydantic models that have constraint keywords (like Field(min_length=1)), the rec_strict_json_schema() function was raising a ValueError because it only considered str and bool as valid terminal types, but JSON Schema constraint keywords have numeric values (int or float).

Example that failed:

from pydantic import BaseModel, Field
from mistralai.extra.utils.response_format import response_format_from_pydantic_model

class ModelWithConstraints(BaseModel):
    items: list[int] = Field(min_length=1)
    name: str = Field(min_length=1, max_length=100)

# This raised ValueError: Unexpected type: 1
response_format = response_format_from_pydantic_model(ModelWithConstraints)

Solution

Added int and float to the type check in rec_strict_json_schema():

if isinstance(schema_node, (str, bool, int, float)) or schema_node is None:
    return schema_node

Test plan

• Added test for rec_strict_json_schema with numeric constraints (minItems, maxItems, minLength, maxLength, minimum, maximum, multipleOf)
• Added test for response_format_from_pydantic_model with constrained Pydantic models
• Added test to verify invalid types still raise ValueError

Fixes #300

🤖 Generated with Claude Code