Add comprehensive tests for KDE vs GMM likelihoods

Introduces three new test modules for clusterless likelihoods: (1) test_gmm_kde_convergence.py verifies GMM convergence to KDE and mathematical consistency, (2) test_kde_gmm_comparison.py provides end-to-end and API compatibility tests for both implementations, and (3) test_kde_gmm_numerical_comparison.py performs detailed numerical and statistical comparisons of KDE and GMM outputs. Also updates test_clusterless_gmm_optimization.py.

Author: edeno