Use a larger time step for MPAS forecast

[liujake]

I believe we are still using 6dx for the forecast time step in our workflow and experiments.
I suggest we try 8dx, in the cycling and longer-range forecasts.
This will make experiment run faster and save a not small % of core hours.
Once we confirm this larger time step works well for you, we should set this as the default in the workflow.

[RGNystrom]

8dx appears to fail for 4d applications because the output times are not evenly divisible by 8dx. I am testing 7.5dx now which I believe should work.