fix(examples): preserve assistant content across streamed tool rounds

[roli-lpci]

The GPT-OSS streaming tool example prints assistant content chunks but never adds them to the content accumulator. When a response includes text followed by a tool call, the next request therefore replays an assistant message with empty content.

This accumulates the streamed content before storing the assistant message. A deterministic test runs the example with a mocked two-round stream and checks that the next request preserves both text chunks before the tool result. No model, credentials, or network requests are needed by the test.

Validation:

• The regression fails against the unchanged example ('' instead of Checking London.) and passes with the fix.
• uv run --python 3.11 --with pytest --with pytest-httpserver --with pytest-anyio python -m pytest -q ollama tests: 103 passed.
• ruff check . and git diff --check: passed.
• Independent local diff review: no findings.

Prepared with AI assistance.