Clean up composition evaluation and add PCA baseline

[nkotelevskii]

Summary

• refactor multidimensional composition evaluation so EntropicOT and baseline aggregators share score loading, matrix stacking, and result-row creation
• add PCAUncertaintyOrdering as a z-scored one-component PCA baseline for every configured composition
• add --skip_pca_baseline for runs that should keep only the previous EntropicOT composition results
• repair scripts/compose_multidimensional_scores.py for the current EntropicOT API
• expand README with the current pipeline and add fast tests for PCA, 1D EntropicOT ordering, and evaluator smoke coverage

Tests

• PYTHONDONTWRITEBYTECODE=1 .venv/bin/python -m unittest discover
• .venv/bin/python -m py_compile scripts/full_evaluation.py scripts/compute_measures_1d.py scripts/compose_multidimensional_scores.py
• git diff --check

[nkotelevskii]

.