Resolves #809

Summary

Adds a public API to retrieve feature importance directly from AutoML:

• automl.get_feature_importance(model="best", kind="raw")
• Supports:
  • model="best" (default)
  • model="all"
  • specific model name
• Supports:
  • kind="raw"
  • kind="normalized" (per-model min-max normalization)

What Changed

• Added AutoML.get_feature_importance(...) in supervised/automl.py.
• Implemented backend logic in BaseAutoML:
  • load and aggregate feature importance from existing *_importance.csv files,
  • return empty DataFrame for models without available importance.
• Added best-effort recompute when importance files are missing:
  • recompute uses permutation importance generation for model learners,
  • if recompute still cannot produce valid importance, returns empty output (no exception).
• Added tests for API behavior in:
  • tests/tests_automl/test_feature_importance_api.py

Notes

• Normalized importance uses per-model min-max scaling to [0, 1].