fix(items): empty list/tuple tool output dropped via all([]) == True

[vidigoat]

Problem

ItemHelpers._convert_tool_output (in src/agents/items.py, reached via the public ItemHelpers.tool_call_output_item) decides whether a tool's list/tuple return value is structured output or should be stringified:

if isinstance(output, list | tuple):
    maybe_converted_output_list = [
        cls._maybe_get_output_as_structured_function_output(item) for item in output
    ]
    if all(maybe_converted_output_list):   # <-- all([]) is True
        return [ ... structured ... ]
    else:
        return str(output)

For an empty list or tuple, the comprehension yields [] and all([]) == True, so the empty collection wrongly takes the structured-output branch and returns an empty ResponseFunctionCallOutputItemListParam ([]). Sent to the Responses API as a function_call_output, an empty structured list drops the tool result entirely — the model sees no output for that tool call — instead of the expected stringified "[]" / "()".

A non-empty plain list like [1, 2, 3] correctly stringifies to "[1, 2, 3]"; only the empty case misbehaves.

Fix

if maybe_converted_output_list and all(maybe_converted_output_list):

Empty collections now fall through to str(output).

Verification

before:  pytest tests/test_tool_output_conversion.py -k empty_list_not_converted or empty_tuple_not_converted
         2 failed  (isinstance([], str) is False -- output was [] not '[]')
after :  26 passed (24 existing + 2 new regression tests)
full suite green:  tests/test_items_helpers.py tests/test_tool_output_conversion.py
                   tests/test_tool_origin.py tests/test_output_tool.py  -> 86 passed

ruff check, ruff format --check, and mypy all pass on the changed files.

Adds test_tool_call_output_item_empty_list_not_converted and ..._empty_tuple_not_converted mirroring the existing "not_converted" test style.