Add perfect cluster diagnostic metrics

[ryux1]

Adds separate perfectly_homogeneous and perfectly_inclusive tracking metrics so failures of the existing perfect efficiency can be attributed to fragmentation or contamination.

The predicates reuse the two halves of perfect_match, preserving the existing 0.99 homogeneity threshold. Both diagnostics use the same particle denominator as perfect; homogeneous matches count unique majority particle IDs so splitting one particle into multiple pure clusters cannot produce an efficiency above 1.

The focused regression covers both directions:

• fragmented pure clusters are homogeneous but not inclusive
• one contaminated cluster containing all hits of its majority particle is inclusive but not homogeneous

Closes #500

Validation:

• GITHUB_ACTIONS=1 PYTHONPATH=src .venv/bin/python -m pytest tests/test_cluster_metrics.py -q (16 passed, 1 repository-declared xfail)
• Ruff 0.4.3 checks passed
• git diff --check