Chaotic bounce results with LockRotation off, implying issue with collision resolver

[Hexicube]

Describe the bug

Bounces at slightly different speeds produce wildly different results in a simple scene, implying an issue with collision resolution.

Before bounce:

After bounce:

To Reproduce

Add this to the root Node3D:

[Export]
public PhysicsMaterial BallMat;

RigidBody3D[] ballClones = new RigidBody3D[10];
override public void _PhysicsProcess(double delta) {
    if (ballClones[0] == null) {
        for (int a = 0; a < 10; a++) {
            ballClones[a] = new RigidBody3D();
            //ballClones[a].LockRotation = true;
            ballClones[a].ContinuousCd = true;
            ballClones[a].LinearDampMode = RigidBody3D.DampMode.Replace;
            ballClones[a].PhysicsMaterialOverride = BallMat;
            ballClones[a].CollisionMask = 0;
            ballClones[a].Position = new Vector3(0, 10f, 0);
            ballClones[a].LinearVelocity = new Vector3(0, a-15f, 0);

            CollisionShape3D shape = new CollisionShape3D();
            SphereShape3D sphere = new SphereShape3D();
            sphere.Radius = 0.1f;
            shape.Shape = sphere;
            ballClones[a].AddChild(shape);

            MeshInstance3D mesh = new MeshInstance3D();
            SphereMesh sphereMesh = new SphereMesh();
            sphereMesh.Radius = 0.1f;
            sphereMesh.Height = 0.2f;
            mesh.Mesh = sphereMesh;
            ballClones[a].AddChild(mesh);

            AddChild(ballClones[a]);
        }
    }
}

Passed in BallMat is 0.0 friction and 1.0 bounce, floor is 0.2 rough friction and 0.5 absorbent bounce.

Place a slightly sloped flat panel (approx 15deg) so that the bounce is towards -X (doubt direction/angle matters) at Y=0. Observe that the resulting bounces are chaotic.

Uncomment the LockRotation line. Observe that the resulting bounces are far less chaotic.

This may be due to not implementing ContinuousCd, results are identical with or without it enabled.
Results are also far less chaotic when changing physics ticks per second from 60 to 1000.

Issue vanishes with a frictionless bounce, which may be because zero friction bounces do not impart any spin on the ball.
I suspect the added spin is causing one tick of friction calculation with varied levels of penetration and spin.

Expected behavior

Bounce should be far more consistent, like how it is with LockRotation enabled:

Note: This is a consistent but wrong bounce, should be far more to the right.
Fully expected bounce looks like this (set friction to 0):

Environment:

• OS: Manjaro
• Version: 0.8.10
• Godot Version: 4.5
• Type: 3D single

[Ughuuu]

Not quite sure what to make of this issue, it's possible there is an issue with rapier about this? Or the the object is not configured correctly with ccd?

[Hexicube]

The CCD option does nothing, it's possible that the Godot wrapper isn't passing that on.

Either way, collisions with CCD off are inconsistent/wrong because friction is applied on a bounce far more than I would intuitively expect it to be. This is the more pressing half of the problem for me since working CCD would likely not resolve this.