Enable executing atm_bdy_set_scalars_work on gpus

[jim-p-w]

This PR enables executing the atm_bdy_set_scalars_work subroutine on GPUs.
This is accomplished using OpenACC directives.

Tested with a regional test case.
Baseline results were obtained from building the develop branch with:
make -j4 nvhpc CORE=atmosphere PRECISION=single OPENACC=true
Then the changes in this PR were made and compiled in the same way.

Comparing the results stored in the restart.*.nc file showed no changes.

[abishekg7]

Wondering if we also need a timer start/stop around the call to atm_bdy_set_scalars_work. Other than that everything looks good!

[gdicker1]

Wondering if we also need a timer start/stop around the call to atm_bdy_set_scalars_work. Other than that everything looks good!

I'm wondering the same, but that would probably be another PR. @mgduda what are your thoughts about these timers? Maybe we don't need parent timers for the _bdy_ routines since theses [ACCxfer] timers are meant to be transient?

[mgduda]

The [ACC_data_xfer] timers are intended to be transient, and I expect we'll be removing them when we remove the data transfers for individual routines. If the atm_bdy_set_scalars_work routine doesn't already have a timer, then I suppose the atm_bdy_set_scalars_work [ACC_data_xfer] has less value. Perhaps for now, we can just include the [ACC_data_xfer] timer (especially since it's already been added).

@jim-p-w The code changes themselves look fine to me. Can you modify the commit message:

1. "acc" -> "OpenACC"
2. "gpus" -> "GPUs"
3. note that the commit includes a "atm_bdy_set_scalars_work [ACC_data_xfer]" timer although there is no corresponding "atm_bdy_set_scalars_work" timer that includes the computational work of the routine.