Add mechanism to stop simulation when future divergence is detected

[epernod]

would be even better to be able to avoid the divergence withtout stoping the simulation:

for example unactivate collision or adding damping / more iteration to the solver