Contact simulation fails if bodies are in contact with zero contact force

[amgebauer]

Description

I suspect that if we compute the deformation gradient too exact (F=I, instead of ≈I [~1e-17], #1565), it might be that the contact set is active, but the contribution to stiffness is zero causing the linear solver to fail. The problematic tests are here, here, here and here. Note: In newer revision, these tests are designed that this singularity is circumvented.

Possible Solution and Definition of Done

• Fix singularity on contact algorithm side.