Bare object schemas in tool parameters silently strip all keys (empty create_model + pydantic extra='ignore')

[spichen]

Bug

pyagentspec/adapters/_utils.py::_build_type_from_schema converts an object schema with no declared properties — a bare {"type": "object"}, e.g. as items of an array parameter — into a pydantic model built by create_model() with zero fields. Pydantic v2 defaults to extra="ignore", so validating the LLM's arguments against that model silently drops every key: the tool receives {} (or [{}, ...] for arrays of objects) regardless of what the model actually sent, with no error anywhere.

Reproduction

from pyagentspec.adapters._utils import create_pydantic_model_from_properties
from pyagentspec.property import Property

model = create_pydantic_model_from_properties(
    "ToolArgs",
    [Property(title="components", json_schema={"type": "array", "items": {"type": "object"}})],
)
parsed = model(components=[{"id": "root", "component": "Card"}])
print(parsed.components)  # [ComponentsItem()] — id/component are gone

Downstream symptom: any tool whose parameter schema uses opaque objects fails input-dependent logic with confusing errors, because the arguments arrive empty no matter what the model sends.

Expected

Per JSON Schema semantics, an object with no property constraints accepts any object. A bare object schema (no properties, additionalProperties not false) should map to a passthrough Dict[str, Any] rather than a stripping empty model. Schemas that declare properties/required or set additionalProperties: false should keep building typed models as today.

I have a fix with a regression test ready and will open a PR referencing this issue.