fix(ollama): avoid UnboundLocalError on empty model stream

[he-yufeng]

Description

OllamaModel.stream reads event after the async for event in response loop to build the message_stop and metadata chunks. When the model returns an empty stream (no chunks at all), event is never bound, so event.done_reason raises UnboundLocalError instead of closing the turn cleanly.

Repro against current main:

async def empty():
    if False:
        yield None

# client.chat returns an empty async iterator
[x async for x in model.stream(messages)]
# UnboundLocalError: cannot access local variable 'event'

Fix:

• initialize event = None before the loop
• fall back to None for the stop reason when there was no event (format_chunk already maps anything that is not tool_use/length to end_turn)
• skip the metadata chunk when no event arrived, since there is no usage to report

This mirrors what GeminiModel.stream already does for the empty-stream case, so the providers stay consistent. An empty stream now ends with a normal messageStop(end_turn) and no metadata chunk.

Related Issues

None tracked; found while auditing the model providers for the empty-stream edge case.

Documentation PR

Not needed, no public API or behavior change beyond not raising.

Type of Change

Bug fix

Testing

Added test_stream_empty_response, which fails on main with the UnboundLocalError above and passes with the fix.

python -m pytest tests/strands/models/test_ollama.py
# 34 passed
ruff check / ruff format --check  # clean
mypy src/strands/models/ollama.py # Success: no issues found

Checklist

• I have read the CONTRIBUTING document
• I have reviewed and understand every line of code in this PR, including any generated by AI tools, and I can explain why it works
• My change is focused and reasonably small; I have split unrelated work into separate PRs
• I have added any necessary tests that prove my fix is effective or my feature works
• My changes generate no new warnings