Description

Fixes #3867

When a tool returns list[dict], the client deserializes each dict as a Root() dataclass with no fields instead of preserving the original dict data:

@server.tool
def get_temperatures() -> list[dict]:
    return [{"city": "NYC", "temp": 72}, {"city": "LA", "temp": 85}]

# Before fix:  result.data == [Root(), Root()]
# After fix:   result.data == [{"city": "NYC", "temp": 72}, {"city": "LA", "temp": 85}]

Root cause

_get_from_type_handler's "object" branch always fell through to _create_dataclass for schemas without properties (like {"type": "object", "additionalProperties": true} — the JSON Schema for bare dict). This created an empty dataclass named Root.

The top-level json_schema_to_type already handled this correctly — returning dict[str, Any] — but that logic was not reachable from _schema_to_type, which is used for nested schemas (e.g., array items).

Fix

Extract _object_schema_to_type to unify the four object-schema cases so both top-level and nested paths produce the correct type:

1. No properties + additionalProperties — dict[str, T]
2. No properties + no additionalProperties — dict[str, Any]
3. properties + additionalProperties is True — Pydantic BaseModel
4. properties only — dataclass

Contribution type

• Bug fix (simple, well-scoped fix for a clearly broken behavior)
• Documentation improvement
• Enhancement

Checklist

• This PR addresses an existing issue (list[dict] return type annotation becomes Root() in testing #3867)
• I have read CONTRIBUTING.md
• I have self-reviewed my changes
• Tests included: unit tests (schema roundtrip) + integration test (client call_tool)
• All 157 json_schema_type tests pass
• All 2548 server tests pass
• All 294 tool tests pass
• ruff passes clean