Deterministic simulation for concave and compound collision shapes: added childShapeA/B to b3Contact4 + sort on them

Use tetrahedra instead of barrel for convex demo (until performance for edge-edge is improved)
Increased #overlapping pair capacity from 12 to 16 / objec

src/Bullet3OpenCL/RigidBody/kernels/batchingKernels.h

@@ -75,6 +75,12 @@ static const char* batchingKernelsCL= \
 "\n"
 "	int m_bodyA;//sign bit set for fixed objects\n"
 "	int m_bodyB;\n"
+"\n"
+"	int	m_childIndexA;\n"
+"	int	m_childIndexB;\n"
+"	int m_unused1;\n"
+"	int m_unused2;\n"
+"\n"
 "}Contact4;\n"
 "\n"
 "typedef struct \n"