"Pulling" problems only run with smallest basis size for OpInf-based couplings

[ikalash]

There seems to be some weird behavior in Norma when running OpInf for the "pulling" problems (notched cylinder, tension specimen, cylinder, cuboid), described below for the specific case of the linear elastic cuboid problem:

• With the combined basis approach, the energy-based 2-mode basis works great. All larger basis sizes lead to the blow up of |residual|.
• With the split basis approach, all basis sizes, even the energy-based one (3 modes) lead to blow up of |residual|.

I am pretty sure the second case did not happen with the 3 mode basis before when I ran it. The learned operators K become horribly conditioned with increasing basis size.