The normal traction and opening are still quite small, so I would try increasing the zero_tolerance_normal to try to keep the fault closed. Sometimes small asymmetries result in very small amounts of fault opening that are difficult to avoid.
See examples/2d/subduction/step06.cfg for an example of how to set zero_tolerance_normal.
I would use something like
[pylithapp.timedependent.interfaces.fault]
# Zero tolerance for locking fault (no slip). Must be larger than ksp tolerance.
zero_tolerance = 1.0e-8
zero_tolerance_normal = 1.0e-6