It looks like someone (Ben?) moved the NOX configure test to be nested
inside of the configure test for Trilinos.
The reason I didn't do this initially is that it's actually possible
to use NOX without the rest of Trilinos. In fact, you can even use
NOX with Petsc Vectors! Currently that's not possible with the way
I've implemented NoxNonlinearSolver in libMesh... but I left the
possibility open for configuring the library with NOX support and
_without_ the rest of Trilinos hoping that eventually we could try to
use NOX with Petsc vectors (NOX, in fact, supports many different
numeric data structure implementations).
I'm not opposed to leaving it nested for now (since it _is_ required
to have Epetra to use NoxNonlinearSolver for now)... but I just
thought I would provide some insight into why I didn't do that in the