inaccurate collisions using with GImpact shapes

[stephengold]

If scaling is applied to a collision object with a GImpact shape, it no longer collides properly with other physics objects. This issue was discovered by @louhy recently, in master branch, with both jme3-jbullet and jme3-bullet.

Enabling CCD and increasing solverNumIterations didn't seem to help.

On the web, I found a programmer's guide for GImpact:
http://gimpact.sourceforge.net/reference_html/programmers_guide.html
It says:
"At end, you must call btGImpactMeshShape.updateBound for ensure that the shape will build its internal Box set structure"

Currently jme3-bullet-native calls updateBound() during createShape(long) but not during setScale(Vector3f). Thus, scaling results in a size mismatch between the collision shape and its box set structure in GImpact. The more you scale up the collision shape, the worse the mismatch becomes.

[stephengold]

Oops! I forgot @louhy already opened an issue for this: #1120.