FIX: start event loop thread when fit_predict_async called outside MCP server context

[echo-xiao]

Reference Issues/PRs

Fixes #446

What does this implement/fix? Explain your changes.

• Fixed background async training job stuck at 0% progress when fit_predict_async_tool is called outside the MCP server context. The event loop exists but is not running, so scheduled coroutines never execute. Fix starts a background thread to run the event loop when it is not already running.

• Fixed KeyError: 'data_source' in background_training_example.py by supporting both 'data_source' and 'dataset' keys for compatibility between PyPI and source versions.

Does your contribution introduce a new dependency? If yes, which one?

No. Uses only Python's built-in threading and asyncio modules.

What should a reviewer concentrate their feedback on?

• The threading fix in fit_predict.py: whether starting a daemon thread is the correct approach when the event loop is not running
• Whether this fix is safe to use within the MCP server context as well

Any other comments?

This bug was discovered while running the example scripts directly with python examples/background_training_example.py. The root cause is that fit_predict_async_tool uses run_coroutine_threadsafe(), which requires a running event loop. When called outside the MCP server, the loop exists but is not running, so coroutines are scheduled but never executed.

PR checklist

For all contributions

• I've added unit tests and made sure they pass locally (make check).
• I've added the tool to the online documentation in docs/source/.
• I've updated the existing example scripts or provided a new one to showcase how my tool works in examples/.