fix(responses): include text.format when verbosity set for plain-text responses

When model_settings.verbosity was set and no output schema was provided,
the code previously built response_format = {"verbosity": ...} which
is an invalid `text` payload for the Responses API. The Responses API
expects a `text` object to include a `format` field (for plain text use
"format": "text").

This patch ensures a valid payload is sent by using:
  {"format": "text", "verbosity": <value>}

Impact:
- Prevents 400/BadRequest errors when callers use verbosity with plain text
  responses (no JSON schema).
- Backwards compatible for callers that already use output_schema.

Author: mshsheikh

src/agents/models/openai_responses.py

@@ -299,7 +299,9 @@ async def _fetch_response(
             if response_format is not omit:
                 response_format["verbosity"] = model_settings.verbosity  # type: ignore [index]
             else:
-                response_format = {"verbosity": model_settings.verbosity}
+                # When no output_schema is present, text needs an explicit format key.
+                # Use the plain text format with verbosity to produce a valid `text` payload.
+                response_format = {"format": "text", "verbosity": model_settings.verbosity}
 
         stream_param: Literal[True] | Omit = True if stream else omit
 
@@ -501,7 +503,7 @@ def _convert_tool(cls, tool: Tool) -> tuple[ToolParam, ResponseIncludable | None
             }
             includes = None
         else:
-            raise UserError(f"Unknown tool type: {type(tool)}, tool")
+            raise UserError(f"Unknown tool type: {type(tool)} for tool {tool!r}")
 
         return converted_tool, includes