nested solvers should be able to submit only F(x) "in parallel"

[mmckerns]

nested solvers should be able to submit only F(x) "in parallel", instead of submitting N optimizations.

[mmckerns]

Note to Michael Aivazis:

You showed two design slides for the UQ pipeline,
one where N optimizations are launched in parallel
on N nodes, and one where only the 2N simulations
are launched on the N nodes.

There's an obvious difference, in that the optimizer
itself does not need to be launched on the nodes
with the simulations.  The good thing is that for
several of the global optimizers, that is already the
case.  The DifferentialEvolutionSolver and the
ParticleSwarmSolver only launch one instance of
themselves locally, while they launch 2N simulations.

The particular pipeline as configured today, however,
uses the BatchGridSolver, which *does* launch
the N optimizers on the remote nodes.  Only the two
pseudo-global (i.e "parallel batch") solvers do this,
and it's an easy fix to move the interface inside of
the solvers so that it launches similarly to the other
solvers.  This was an oversight on my part for the
two batch solvers that I developed, but the fix is
easy and I'll make the change in the solver...