Description

Add SDF-based mesh-mesh collision detection with contact reduction.

• Mesh-mesh collisions: New SDF-based collision detection between mesh shapes. Each mesh gets a sparse (narrow-band) + coarse SDF volume computed at model finalization.
• Contact reduction: Shared-memory GPU algorithm reduces contacts per shape pair using icosahedron-based normal binning (20 bins × 6 spatial directions × configurable beta weights). Applies to mesh-plane and mesh-mesh collisions.
• SDF extrapolation: Handles queries outside the narrow band (fallback to coarse grid) and outside the extent (linear extrapolation from boundary).
• Performance: Uses tile scans and atomic operations to minimize synchronization overhead. Configurable via reduce_contacts parameter in CollisionPipelineUnified.

Newton Migration Guide

Please ensure the migration guide for warp.sim users is up-to-date with the changes made in this PR.

• The migration guide in docs/migration.rst is up-to date

Before your PR is "Ready for review"

• Necessary tests have been added and new examples are tested (see newton/tests/test_examples.py)
• Documentation is up-to-date
• Code passes formatting and linting checks with pre-commit run -a

Summary by CodeRabbit

• New Features

  • Optional contact‑reduction pipeline and SDF‑driven narrow‑phase for mesh–mesh and mesh–plane collisions; per‑shape SDF data, model/build flags to enable mesh‑SDF collisions, and runtime toggle to reduce contacts.

• Refactor

  • More consistent geometry type handling and per‑thread buffering for robust collision routing and triangle collection.

• Tests

  • Comprehensive SDF, mesh‑SDF compute, and contact‑reduction test suites (CPU/CUDA).

• Chores

  • Relaxed bunny rest‑state tolerance; improved handling for structured arrays and recorder serialization.

_{✏️ Tip: You can customize this high-level summary in your review settings.}