fix(litellm): record partial output on early stream close

[px12811]

Fixes #4011

Summary

LiteLLM streaming spans only recorded output.value and llm.output_messages after the stream loop completed. If an application stopped iterating early and closed or dropped the stream, the generator was closed at its yield and the span was ended with no output attributes.

This change extracts the streaming output finalization into a shared helper. Normal completion records the full output as before, while generator shutdown also records the output received so far before ending the span.

Validation

• Added sync and async regression tests that consume one chunk, close the stream, and assert both output.value and llm.output_messages.0.message.content
• tox run -e py313-ci-litellm on macOS: 131 passed
• Focused tests/test_streaming_transparency.py: 8 passed on Python 3.10, 3.13, and 3.14
• Ruff format, Ruff lint, and mypy passed

[github-actions]

CLA Assistant Lite bot All contributors have signed the CLA ✍️ ✅

[px12811]

I have read the CLA Document and I hereby sign the CLA