Add VBD and AVBD paper-parity foundation

[jslee02]

Summary

• Establishes the fail-closed DART-owned VBD/AVBD paper-parity foundation for
  DART 7, including public solver selection, rigid pair constraints, and
  independently attributable Sequential Impulse, fixed-penalty VBD, and AVBD
  Figure 13 rows.
• Adds machine-checked 88-row VBD and 88-row AVBD contracts. All 176 rows remain
  partial or missing; this PR does not claim paper parity, GPU parity, or a
  paper/reference speedup.

Motivation / Problem

• PLAN-104 needs one shared, reviewable foundation before the remaining work
  can be completed in two ordered PRs: VBD first, then AVBD. Existing partial
  implementations and evidence did not fail closed on solver identity,
  source/build provenance, visual outcomes, CPU/CUDA coverage, or comparable
  achieved-accuracy performance.

Changes / Key Changes

• Adds canonical paper/site/video/demo inventories and a validator for 176
  VBD/AVBD completion predicates.
• Implements AVBD quasi-Newton geometric stiffness, deterministic parallel
  dual/stiffness updates, articulated finite rows and motors, physical fracture
  accounting, and public C++/dartpy rigid-solver configuration.
• Adds a public fixed-penalty VBD comparison path and makes the default
  Sequential Impulse family own hard rigid pair rows, bounded velocity motors,
  fracture, and post-stabilization rather than borrowing AVBD projection.
• Adds source-bound Figure 13 captures, semantic review records, packet-chain
  validation, and descriptive same-host benchmark evidence for SI, VBD, and
  AVBD.
• Updates PLAN-104 and its dev-task handoff with the exact publication state
  and the required foundation -> VBD PR -> AVBD PR execution sequence.

Before / After

Before	After
Partial VBD/AVBD mechanisms and demo packets had no aggregate fail-closed completion authority.	Two machine-checked 88-row contracts make solver identity, CPU/CUDA, visual, and performance gaps explicit.
Public hard rigid pair rows borrowed AVBD projection state.	Sequential Impulse owns hard pair rows; finite-stiffness rows fail closed to VBD/AVBD.
Figure 13 rows were not independently attributable or source-bound as a three-method comparison.	SI, fixed-penalty VBD, and AVBD have distinct partial outcome oracles and one source-bound packet chain.

Testing

Current head 57bce9a5cedf:

• DART_DISABLE_COMPILER_CACHE=ON pixi run test-all (all 6 phases passed;
  lint, release/debug builds, 229/229 core C++ tests, 81 runnable simulation
  tests, the full Python corpus, and documentation)
• PYTHONUNBUFFERED=1 DART_DISABLE_COMPILER_CACHE=ON pixi run -e cuda test-all
  (all 7 phases passed; 213/213 core C++ tests, 80 runnable simulation tests,
  the full Python corpus, 8/8 CUDA device tests, and all 9 benchmark-smoke
  commands)
• build/default/cpp/Release/bin/test_world --gtest_filter=World.PublicVbdFamilyRequiresFinitePointJointProjectionStiffness
  (1/1 public fixed/revolute/prismatic/spherical hard-row regression passed)
• pixi run check-avbd-packets (58 packets)
• pixi run check-plan104-paper-parity (176 canonical rows)
• Fresh six-capture source digest:
  7a96ffd3b73b9649961ed858f7d9967044b6e2fe47477227e04e2840c5f0b4d0

Visual verification

• The foundation has six source-bound 1280x720 Figure 13 stills and three
  separate semantic reviews for SI, fixed-penalty VBD, and AVBD. They pass
  engine ViewReports, pixel-integrity checks, image verdicts, and
  original-resolution review against the pinned paper page.
• These are partial Figure 13 checkpoints, not full VBD or AVBD paper
  replication. No video bundle was generated for this foundation slice, and
  transient media is intentionally not committed.
• The six current stills can be uploaded to this PR as GitHub attachments
  without another code PR. Full paper/site/official-video/source-demo image,
  GIF, and video evidence will ship inside exactly two follow-ups: the VBD
  completion PR first, then the AVBD completion PR. There will be no separate
  media-only PR.
• This follows the repository-hygiene boundary demonstrated by Add opt-in exact-Coulomb FBF friction solver and fix split-impulse impulse loss #3377: generated
  media stays ignored locally and reviewer-facing media uses GitHub
  attachments. It does not copy Add opt-in exact-Coulomb FBF friction solver and fix split-impulse impulse loss #3377's still-incomplete claim boundary.

Breaking Changes

• None
• No public source-compatibility break. Experimental DART 7 rigid-solver
  behavior now attributes hard pair constraints to Sequential Impulse and
  routes finite-stiffness pair constraints to VBD/AVBD.

Related Issues / PRs (backports)

• PLAN-104
• Media-hygiene reference: Add opt-in exact-Coulomb FBF friction solver and fix split-impulse impulse loss #3377
• Main-only DART 7 work; no DART 6 backport is planned.

---

Checklist

• Milestone set (DART 7.0)
• CHANGELOG.md updated per docs/onboarding/changelog.md
• Add unit tests for new functionality
• Document new methods and classes
• Add Python bindings (dartpy) if applicable

[jslee02]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Hooray!

Reviewed commit: 404d66b98d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 995be410b9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[jslee02]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 58bf1c3e85

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[jslee02]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 57bce9a5ce

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[codecov]

Codecov Report

❌ Patch coverage is 83.58779% with 301 lines in your changes missing coverage. Please review.
✅ Project coverage is 82.00%. Comparing base (ef10cb2) to head (57bce9a).

Files with missing lines	Patch %	Lines
...simulation/detail/deformable_vbd/block_descent.hpp	39.28%	119 Missing ⚠️
...imulation/detail/rigid_avbd/rigid_block_kernel.hpp	73.07%	42 Missing ⚠️
...mulation/detail/rigid_avbd/rigid_world_contact.hpp	83.41%	32 Missing ⚠️
...rt/simulation/compute/rigid_body_contact_stage.cpp	94.59%	29 Missing ⚠️
dart/simulation/detail/rigid_pair_constraint.hpp	88.73%	24 Missing ⚠️
...art/simulation/compute/variational_integration.cpp	87.57%	22 Missing ⚠️
dart/simulation/compute/parallel_executor.cpp	86.86%	13 Missing ⚠️
dart/simulation/world.cpp	95.32%	10 Missing ⚠️
...ion/detail/deformable_vbd/quasi_newton_hessian.hpp	64.28%	5 Missing ⚠️
...tion/detail/deformable_vbd/parallel_row_update.hpp	72.72%	3 Missing ⚠️
... and 2 more

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3432      +/-   ##
==========================================
+ Coverage   81.67%   82.00%   +0.33%     
==========================================
  Files         667      670       +3     
  Lines       92082    93365    +1283     
==========================================
+ Hits        75209    76565    +1356     
+ Misses      16873    16800      -73     

Flag	Coverage Δ
unittests	82.00% <83.58%> (+0.33%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
dart/simulation/comps/joint.hpp	100.00% <ø> (ø)
dart/simulation/comps/rigid_body.hpp	100.00% <ø> (ø)
...on/compute/detail/deformable_avbd_replay_state.hpp	100.00% <100.00%> (ø)
...rt/simulation/compute/detail/world_step_stages.hpp	100.00% <ø> (ø)
...art/simulation/compute/variational_integration.hpp	95.65% <ø> (ø)
dart/simulation/detail/world_step_schedule.hpp	98.18% <100.00%> (+0.03%)	⬆️
dart/simulation/io/binary_io.hpp	100.00% <ø> (ø)
dart/simulation/world.hpp	100.00% <ø> (ø)
...t/simulation/compute/deformable_dynamics_stage.cpp	86.30% <66.66%> (-0.01%)	⬇️
.../detail/deformable_vbd/finite_stiffness_kernel.hpp	56.41% <91.66%> (+11.58%)	⬆️
... and 10 more

... and 5 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.