Improving RBE accuracy for static problems

[timryanb]

The RBE elements use a Lagrange Multiplier formulation to compute their stiffnesses. This approach leads to zeros being present in some diagonals of the stiffness matrix, which causes issues when the preconditioner is factored. To get around this, the elements add a small artificial term to their diagonals. This fixes the PC issue, but adds error to the Jacobian.

This PR:

• Separates out the artificial stiffness from the true Jacobian and only adding it in for the preconditioner refactor step
• Adds options to allow user to set the scaling factors used in the RBE elements through the pytacs problem classes
• Improves the default rel. convergence tolerance of static problems featuring RBE elements from ~1e-7 to ~1e-13