Fix 3D Physics bugs caused by scaling game objects.

[a-daniel]

The problem:
In 3D Physics, scaling game objects causes problems in particular cases:

1. Scale > 1 and collision shape position != (0,0,0) => collision object might fly away on contact
2. Scale > 1 and collision shape is capsule => collision object bounces really fast on contact
3. Scale < 1 and collision shape is capsule => collision detection happens later than expected (requires significant overlap between objects).
4. Collision shape is capsule and scale > 1 and physics scale < 1 => collision shape unexpectedly increases in size .

The solution:
Patched Bullet3D.

Fix #12261
Fix #11246
Fix #11024
Fix #7738
Fix #11308

Technical changes

• In btCompoundShape::setLocalScaling the new scale is applied to the unscaled origin.
• In btCapsuleShape:setLocalScaling, the scale is applied only to shape dimensions (m_implicitShapeDimensions) and m_localScaling is kept as 1. This workaround preserves the correct scale required by broad phase(m_implicitShapeDimensions) but by forcing m_localScaling to 1 it solves the problem of the narrow phase applying the scale two times when computing collision depth.
• In btCapsuleShape the collision margin is not part of m_implicitShapeDimensions and is not used in computation of AABB. This way we avoid bloating the shape size in case of high scale values.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4c8d807c5a

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Preserve absolute scale when updating capsule dimensions

This change applies scaling directly to m_implicitShapeDimensions after removing btConvexInternalShape::setLocalScaling(scaling), which previously normalized scale via absolute(). As a result, negative uniform scales now produce negative capsule radius/half-height, creating invalid geometry and unstable contacts. This is reachable in Defold’s 3D sync path because physics_3d.cpp forwards GetUniformScale() into setLocalScaling(...); mirrored objects (for example scale -1) will now feed negative values into this assignment.

Useful? React with 👍 / 👎.

[CLAassistant]

All committers have signed the CLA.