[BUG] agentic workflow examples drift from current MCP behavior and are not smoke-tested

[biru-codeastromer]

Summary

The repository ships several agentic / MCP workflow examples, but they are not currently exercised by tests. As a result, some example narratives have drifted from the current runtime behavior.

Reproduced drift

• examples/02_llm_query_simulation.py compares models on sunspots, but that dataset is not in the currently exposed demo dataset list, so the simulated fit_predict calls return success: false.
• The same example uses Detrend instead of the actual estimator name Detrender, producing an Unknown estimator validation result.
• examples/01_forecasting_workflow.py uses a forecaster→forecaster invalid-composition demonstration that no longer reflects current validator behavior on main.

Why this matters

These example scripts are exactly the kind of workflow artifacts an LLM/agent user will read and run first. If they drift from the runtime, they become misleading documentation for agentic MCP usage.

Expected behavior

• documented agentic/MCP examples should use dataset names and estimator names that work on current main
• high-level example scripts should be covered by lightweight smoke tests so drift is caught in CI before release