Show feature importance for tree-based models

[jeanclawd]

Problem

Every tree-based model in scikit-learn exposes `feature_importances_`, which is one of the most useful diagnostics a data scientist looks at after training. Scikit-Learner trains all of these models but never surfaces the information.

Proposal

When the user trains any tree-based model (Decision Tree, Extra Tree, Random Forest, Gradient Boosting, HistGradientBoosting, AdaBoost, Bagging — plus XGBoost / LightGBM once #3 lands), render a horizontal Plotly bar chart of feature importances sorted descending.

Implementation notes

• In `frontend/py/learner.py`, in the result dict returned by `train()`, add a `feature_importances` field whenever `hasattr(model, 'feature_importances_')`. Map back to the original feature names.
• In `frontend/js/app.js`, add a "Feature Importance" tab (or sub-section in the existing model detail view) that renders the chart when the field is present.
• Skip silently for non-tree models — no error UI needed.

Acceptance criteria

• Tree-based models show a feature-importance chart immediately after training.
• Bars are sorted descending, labelled with the original column names.
• Non-tree models show no chart and no error.

Why now

3-line addition per model in Python, ~30 lines of JS for the chart. Very high payoff for the effort — users see this and immediately get why "the model thinks this".