Preserve hypertoroidal particle counts and axes

[FlorianPfaff]

Bug

HypertoroidalParticleFilter initialized multidimensional particles with a floating-step arange(...) and then applied .squeeze().

This violates the constructor's particle-count and shape contracts in two concrete cases:

• floating-point rounding can include the nominal endpoint. For n_particles=61 and dim=2, NumPy produces 62 grid values, with the final value equal to 2π; after toroidal wrapping this is a duplicate of zero;
• for n_particles=1 and dim>1, .squeeze() removes the particle axis, so one three-dimensional particle is stored as shape (3,) instead of (1, 3). The shared Dirac constructor then interprets the state as three scalar Diracs and creates three weights.

Fix

Generate the one-dimensional angular grid with exact-count linspace(..., num=n_particles, endpoint=False) for every dimension. For multidimensional filters, tile and transpose that grid without squeezing the batch axis.

Regression coverage

• verify that requesting 61 two-dimensional particles creates state shape (61, 2) and 61 weights;
• verify that one three-dimensional particle preserves state shape (1, 3), one weight, and a three-dimensional point estimate.

Validation

• independently reproduced the old 61-particle grid as 62 values, including the nominal 2π endpoint;
• independently reproduced the singleton construction collapsing (1, 3) to (3,);
• refreshed the branch directly onto current main (f196ec41d1d8681e279936db9f1a0a3abfa276e7);
• branch comparison is 2 commits ahead / 0 behind;
• final diff is limited to 8 changed implementation lines and 13 regression-test additions;
• the prior version of this focused patch passed MegaLinter; GitHub Actions on the refreshed head is authoritative for the full repository matrix.

[github-actions]

✅MegaLinter analysis: Success

Descriptor	Linter	Files	Fixed	Errors	Warnings	Elapsed time
✅ COPYPASTE	jscpd	yes		no	no	19.4s
✅ JSON	prettier	7	0	0	0	1.05s
✅ JSON	v8r	7		0	0	2.14s
✅ MARKDOWN	markdownlint	68	0	0	0	1.5s
✅ MARKDOWN	markdown-table-formatter	68	0	0	0	0.5s
✅ PYTHON	black	1866	87	0	0	69.57s
✅ PYTHON	isort	1866	140	0	0	2.17s
✅ REPOSITORY	betterleaks	yes		no	no	1.83s
✅ REPOSITORY	checkov	yes		no	no	43.08s
✅ REPOSITORY	gitleaks	yes		no	no	14.41s
✅ REPOSITORY	git_diff	yes		no	no	0.15s
✅ REPOSITORY	secretlint	yes		no	no	64.25s
✅ REPOSITORY	syft	yes		no	no	3.83s
✅ REPOSITORY	trivy-sbom	yes		no	no	5.04s
✅ REPOSITORY	trufflehog	yes		no	no	23.93s
✅ YAML	prettier	11	0	0	0	0.6s
✅ YAML	v8r	11		0	0	9.23s
✅ YAML	yamllint	11		0	0	0.41s

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 superseded by #5220. Both PRs fix the particle-count/axis-shape regression in the same constructor, but #5220 also removes the artificial multidimensional diagonal collapse and adds regression coverage for marginal spacing. Keeping both would duplicate the same code path while the newer patch preserves the underlying dependence defect.