Although we had some super-secondary matches in the template library,
we had to rely heavily on our "new-fold" cost functions for building
models for target T0281.  We ended up with several slightly different
models that we have difficulty choosing among.

This is try21-opt2, which has 3 strands and avoids the bulge in the
first strand.  It puts helices on both sides of the sheet.  It is
fairly compatible with the contact prediction, and avoids the
S-twisted (left-handed) crossover, but scores fourth among our models
using the unconstrained cost function.