Add halo exchanges to the driver after reading data

[oelbert]

Currently we don't halo exchange after reading data and assume the exchanges at the start of the dynamics step are sufficient. This may not be the case, should we add halo exchanges in the driver at model starts to be sure?

[FlorianDeconinck]

My understanding of the core is that they are. The halos are not suppose to have any values when the call to step_dynamics happens. A quick way to test that is: spin a baroclinic, grab the state right before step_dynamics, flood the halos with bad values and/or NaN. If the dynamics spins out of whack, we have an issue