Fix overflow in point-set distance norms

[FlorianPfaff]

What changed

• replace the overflow-prone built-in Euclidean pairwise norm with scale-normalized norm evaluation
• keep the correction directly inside src/pyrecest/utils/metrics.py
• add public OSPA and GOSPA regressions for large finite coordinates

Root cause

_pairwise_distances() delegated directly to np.linalg.norm. For a finite difference vector such as [1e308, 1e308], the internal sum of squares overflows even though the Euclidean norm, approximately 1.4142e308, is representable. OSPA and GOSPA then clip the resulting infinity to the cutoff and overstate localization error whenever the cutoff exceeds the true distance.

Impact

Built-in OSPA/GOSPA distances now preserve representable Euclidean distances at extreme finite coordinate scales. Ordinary-scale behavior and custom distance_fn handling are unchanged.

This replaces the valid bug fix attempted in #4951 with the requested focused implementation: no wrapper module, no duplicated metrics implementation, and no import-time monkey patching.

Validation

• reproduced the pre-fix overflow under np.errstate(over="raise", invalid="raise")
• validated both public OSPA and GOSPA calls against np.hypot(1e308, 1e308)
• syntax-compiled the modified source and regression module
• branch is current with main; final diff is limited to one in-place source change and one regression file

GitHub Actions provides the full supported-version and backend validation matrix.

[github-actions]

✅MegaLinter analysis: Success

Descriptor	Linter	Files	Fixed	Errors	Warnings	Elapsed time
✅ COPYPASTE	jscpd	yes		no	no	29.38s
✅ JSON	prettier	7	0	0	0	1.31s
✅ JSON	v8r	7		0	0	3.32s
✅ MARKDOWN	markdownlint	68	0	0	0	2.11s
✅ MARKDOWN	markdown-table-formatter	68	0	0	0	0.71s
✅ PYTHON	black	1808	55	0	0	95.4s
✅ PYTHON	isort	1808	87	0	0	2.8s
✅ REPOSITORY	betterleaks	yes		no	no	2.5s
✅ REPOSITORY	checkov	yes		no	no	54.51s
✅ REPOSITORY	gitleaks	yes		no	no	17.29s
✅ REPOSITORY	git_diff	yes		no	no	0.32s
✅ REPOSITORY	secretlint	yes		no	no	82.77s
✅ REPOSITORY	syft	yes		no	no	6.17s
✅ REPOSITORY	trivy-sbom	yes		no	no	5.58s
✅ REPOSITORY	trufflehog	yes		no	no	31.68s
✅ YAML	prettier	11	0	0	0	0.82s
✅ YAML	v8r	11		0	0	13.07s
✅ YAML	yamllint	11		0	0	0.67s

Notices

📣 MegaLinter 9.5.0 is out! Discover the new features and security recommendations in the release announcement. (Skip this info by defining SECURITY_SUGGESTIONS: false)

See detailed reports in MegaLinter artifacts

Your project could benefit from a custom flavor, which would allow you to run only the linters you need, and thus improve runtime performances. (Skip this info by defining FLAVOR_SUGGESTIONS: false)

• Documentation: Custom Flavors
• Command: npx mega-linter-runner@9.6.0 --custom-flavor-setup --custom-flavor-linters PYTHON_BLACK,PYTHON_ISORT,COPYPASTE_JSCPD,JSON_V8R,JSON_PRETTIER,MARKDOWN_MARKDOWNLINT,MARKDOWN_MARKDOWN_TABLE_FORMATTER,REPOSITORY_CHECKOV,REPOSITORY_GIT_DIFF,REPOSITORY_GITLEAKS,REPOSITORY_BETTERLEAKS,REPOSITORY_SECRETLINT,REPOSITORY_SYFT,REPOSITORY_TRIVY_SBOM,REPOSITORY_TRUFFLEHOG,YAML_PRETTIER,YAML_YAMLLINT,YAML_V8R

Show us your support by starring ⭐ the repository

[FlorianPfaff]

Closing as low value. This changes core OSPA/GOSPA distance evaluation solely for coordinate differences around 1e308. That scale is not a meaningful supported tracking regime, so the special normalization path and regression surface are not justified.