DOC/TEST: add agentic model-selection workflow example

[biru-codeastromer]

Summary

This PR adds a runnable agentic model-selection workflow example using the current MCP tool surface, plus a smoke test to keep it working.

Closes #403.

Why

The repo already exposes the primitives needed for a simple agent loop — discovery, description, instantiation, evaluation, forecasting, and code export — but it did not yet ship a concrete example that ties them together into a truthful end-to-end workflow.

What this adds

• examples/07_agentic_model_selection.py
  • discovers forecasting candidates
  • inspects capabilities
  • instantiates and evaluates multiple candidates
  • ranks them by mean MAPE
  • selects a winner
  • generates a forecast
  • exports reproducible code
• tests/test_agentic_workflow_example.py
  • smoke-tests the workflow so the example stays aligned with current runtime behavior

Verification

• .venv/bin/ruff check examples/07_agentic_model_selection.py tests/test_agentic_workflow_example.py
• .venv/bin/ruff format --check examples/07_agentic_model_selection.py tests/test_agentic_workflow_example.py
• .venv/bin/python examples/07_agentic_model_selection.py
• .venv/bin/pytest tests/test_agentic_workflow_example.py -q
• .venv/bin/pytest -q

[Shashankss1205]

It no longer runs on main: the examples/ directory was removed in #474, and the script imports list_estimators_tool, describe_estimator_tool, instantiate_estimator_tool, evaluate_estimator_tool and fit_predict_tool, all replaced during the tool consolidation (query_registry, describe_component, instantiate, evaluate, fit+predict). The smoke test fails with ModuleNotFoundError. Closing as obsolete along with #403.