Port check diffusive CFL loop in MOM tracer hor diff

[uwagura]

There was a block of code in MOM_tracer_hor_diff that only ran when check_diffusive_CFL was true. This setting is true in benchmark_ALE, but not in benchmark. Since the khdt_* arrays needed for this block of code were set on the device , the host values used for this calculation contained garbage values in benchmark_ALE, which could occasionally cause the num_itts variable set in this block to take on extremely large values and make the benchmark_ALE config appear to hang.

This PR simply ports this block of code so that max_CFL, and by extension num_itts, are always set correctly.

[github-actions]

GPU Port Coverage

Overall: 2046 / 5662 portable executed lines ported (36.1%)
Since base branch: +3 ported lines (+0.0 pp)

Files touched by this PR: 36 / 42 portable executed lines ported (85.7%)

_{Full per-file / per-routine breakdown: see the "gpu-port-report" job summary and artifact.}

Full line-by-line coverage report

[edoyango]

clearly i wasn't disciplined enough with wrapping sections with !$omp target update! thx for persisting with this.

Ofc the hang decided not to show up when i actually want it, but it looks reasonable and reproduces answers.

[edoyango]

Suggested change

	do concurrent (j=js:je, i=is:ie) DO_LOCALITY(reduce(max:max_CFL))
	CFL(i,j) = 2.0*((khdt_x(I-1,j) + khdt_x(I,j)) + &
	(khdt_y(i,J-1) + khdt_y(i,J))) * G%IareaT(i,j)
	if (max_CFL < CFL(i,j)) max_CFL = CFL(i,j)
	enddo ; enddo
	max_CFL = max(max_CFL, CFL(i,j))
	enddo
	do concurrent (j=js:je, i=is:ie) DO_LOCALITY(reduce(max:max_CFL))
	max_CFL = max(max_CFL, 2.0*((khdt_x(I-1,j) + khdt_x(I,j)) + &
	(khdt_y(i,J-1) + khdt_y(i,J))) * G%IareaT(i,j))
	enddo

This is just a suggestion, so you don't have to take it since it's noncritical. CFL (the array) isn't used anywhere else, so it could be removed.