Optimize ACC port of atm_advance_scalars_work:4857

[abishekg7]

This PR introduces optimizations for the OpenACC port of atm_advance_scalars_work:4857. It is a redo of #1476 with pre-processor macros so as to not degrade the CPU performance significantly. The table below lists the timings for a real, global 30km experiment on A100 GPU with the nvhpc.

For nvhpc gpu runs, -gpu=math_uniform is introduced as a build flag to ensure optimizations are bit-identical, and the we report the numbers using NV_ACC_TIME=1. The GPU runs are on 1 Derecho GPU node, using 1 A100 via 1 MPI task.

The numbers reported for the CPU runs with gnu and intel compilers use the newly-added timers local to this region, and are averaged across three runs. The CPU runs use a single derecho CPU node each, fully subscribed to 128 MPI tasks.

Version	GPU kernel time (ms)		CPU timer - gnu		CPU timer - intel25
base	36		1.155987		1.0895
Opt 1 + macro swap the seq and vector collapse(2) loops	26.272		1.152207		1.09589
Opt 2 + macro, fuse all scalar_tend_column loops	23.4		1.127923		1.09153
Opt 3 + macro, scalar_tend_column can be replaced by a scalar local to each gang	22.365		1.14839		1.10588
Opt 4 + macro, remove loops to zero scalar_tend_column	17.57

This work was completed in part at the NCAR/NLR/NOAA Open Hackathon, part of the Open Hackathons program. The authors would like to acknowledge OpenACC-Standard.org for their support.