Optimize transform calculations for unit-scale nodes in synchierarchy

[liamdon]

Summary

• Extends unit-scale optimization by tracking unit-scale status on nodes and ancestors
• Uses an optimized matrix multiply path for unit-scale transforms and fast-paths in world transform composition
• Ensures world/local transforms remain correct when only translation/rotation changes

Changes

Core Functionality

• Add internal flags on GraphNode to track unit-scale status and ancestor unit-scale status:
  • _hasUnitScale
  • _ancestorsHaveUnitScale
• Propagate unit-scale status changes through the subtree when a node's local scale changes or when hierarchy topology changes
• Introduce Mat4.mulAffine2NoScale(lhs, rhs) for unit-scale transforms and integrate a fast-path in mulAffine2 when scale is exactly (1,1,1)
• In world transform calculation, use mulAffine2NoScale when this node and its ancestors are unit-scale to bypass redundant scale multiplications
• Maintain correctness of world and local transforms for scenarios where only translation/rotation changes occur

Performance

• Skip nine scale multiplications in common unit-scale hierarchies
• Reduce matrix multiplications and allocations in the synchronization path

Tests

• Extend unit tests to cover unit-scale optimization and ancestor propagation
• Validate world/local transforms match pre-optimization results for unit-scale trees
• Update existing tests to exercise the unit-scale code path

Dependency Updates

• Bump PlayCanvas dependency to 2.9.0-beta.1 in examples/package-lock.json

Refactor

• Centralize unit-scale handling logic in GraphNode and augment transform utilities with a dedicated unit-scale path

Test plan

• Run unit tests for transform/hierarchy syncing
• Validate world/local transforms for unit-scale trees
• Benchmark performance on sizable unit-scale hierarchies

📎 Task: https://www.terragonlabs.com/task/f9946d6e-7d11-4228-abd1-ea617df67866