Handle nonfinite track observation ids

[FlorianPfaff]

Summary

• Treat non-finite numeric observation IDs as missing values in track-matrix normalization.
• Add regression coverage for np.inf and -np.inf in normalize_track_matrix.

Root Cause

normalize_track_matrix special-cased NaN but allowed numeric infinities to reach int(value), which raises OverflowError instead of normalizing invalid observation IDs consistently.

Validation

• py -3.12 -m pytest -q tests\test_track_evaluation.py tests\test_track_metrics.py
• py -3.12 -m ruff check src tests
• git diff --check