• Closes Type ValidatedToolArgs missing pydantic BaseModel #5134

Problem

When a tool function has exactly one "model-like" parameter (BaseModel, dataclass, TypedDict), the JSON schema sent to the model is the parameter's schema directly — so the model generates the model's fields at the top level instead of inside a redundant wrapper. However, Pydantic validation then returned the BaseModel instance itself rather than a {param_name: instance} dict. This violated the ValidatedToolArgs = dict[str, Any] type contract and broke every downstream consumer that relied on a dict-shaped view:

• Capability hooks: after_tool_validate, wrap_tool_validate, before_tool_execute, after_tool_execute, wrap_tool_execute, on_tool_execute_error, on_tool_validate_error
• ValidatedToolCall.validated_args: dict[str, Any] | None
• AbstractToolset.call_tool(tool_args: dict[str, Any], ...)
• args_validator_func(ctx, **args) — silently broken at runtime, since **base_model_instance isn't a valid mapping unpack

Fix

Wrap the validator's core schema with core_schema.no_info_after_validator_function that produces {single_arg_name: value}. The JSON schema sent to the model is unchanged (pydantic's JSON schema generator delegates to the inner schema for function-after), so the model keeps receiving the BaseModel's fields at the top level — but the validator's output is uniformly dict-shaped, so every downstream consumer sees the same contract.

The now-redundant rewrapping inside FunctionSchema._call_args is removed.

Tests

• test_args_validator_single_base_model_arg confirms args_validator now works with single-BaseModel-arg tools (was silently broken)
• test_hooks_receive_dict_args_for_single_base_model_tool confirms after_tool_validate and wrap_tool_execute hooks both receive dict-shaped args

Checklist

• Any AI generated code has been reviewed line-by-line by the human PR author, who stands by it.
• No breaking changes in accordance with the version policy.
• PR title is fit for the release changelog.