mlframe usage examples

Copy-pasteable recipes per feature. Distinct from the deeper guides in docs/*.md -- those describe design + tradeoffs; these describe "here is the snippet to put in your script".

File	Topic
composite_targets.md	Auto-discovery composite targets: opt-in tiers, reading metadata, Markdown report, kill-switch env var, decision tree
fixed_splits.md	Exact, reusable train/val/test splits: record a run's split, retrain with more history on the same test/val, pin holdouts to date windows

For deeper conceptual material see the docs root:

• docs/baseline_diagnostics_guide.md -- BaselineDiagnostics ablation + init_score baseline
• docs/composite_targets_tutorial.ipynb -- 16-cell TVT walkthrough notebook
• docs/MULTI_OUTPUT.md -- multilabel design notes
• docs/SELECTION_BIAS.md -- PU-learning + drift-correction wrappers
• docs/NUMERICAL_STABILITY_REPORT.md -- numerical edge cases audit
• docs/feature_handling_examples.md -- FeatureHandlingConfig recipes
• docs/DEBUGGING_UPSTREAM_ERRORS.md -- when upstream sklearn / LightGBM raises something cryptic