fix: guard None refusal in ItemHelpers.extract_last_content

[rmotgi1227]

Summary

PR #3394 added an or \"\" guard for ResponseOutputText.text in extract_last_content but left the ResponseOutputRefusal branch unguarded:

# Fixed by #3394 ✓
if isinstance(last_content, ResponseOutputText):
    return last_content.text or ""

# Still unguarded before this PR ✗
elif isinstance(last_content, ResponseOutputRefusal):
    return last_content.refusal        # can be None in practice

ResponseOutputRefusal.refusal is typed as str per the Responses API schema, but the same provider gateways (e.g. LiteLLM) and model_construct paths during streaming that surface None for .text can also surface None for .refusal. Without the guard, callers that rely on the -> str return type of extract_last_content receive None and may crash downstream.

The fix adds the same or "" coercion to the refusal branch, matching the precedent from #3394 for text and extract_refusal (which already uses content_item.refusal or "" at items.py:739).

Test plan

Added two regression tests in tests/utils/test_pretty_print_and_items.py mirroring the structure of the tests added in #3394:

• test_extract_last_content_returns_empty_string_for_none_refusal — verifies the method returns "" (not None) when ResponseOutputRefusal.refusal is None
• test_extract_last_content_returns_refusal_normally — verifies unaffected happy-path behaviour

uv run pytest tests/utils/test_pretty_print_and_items.py tests/test_items_helpers.py -v
# 43 passed

make lint  # All checks passed
make typecheck  # Success: no issues found

Issue number

Discovered while auditing for the same guard-None pattern that #3394 and #3375 fixed.

Checks

• I've added new tests (if relevant)
• I've added/updated the relevant documentation (no doc changes needed)
• I've run make lint and make format
• I've made sure tests pass