dtc/hwrf-physics: merge HWRF saSAS with GFS version, update to a more recent version of ufs-weather-model from EMC develop

[climbfuji]

This PR contains

• develop: merge HWRF version of saSAS with GFS version ufs-community/ufs-weather-model#94, "develop: merge HWRF version of saSAS with GFS version"
• update from EMC develop so that the base for both PRs is the same (after the commit for b4b identical results between IPD and CCPP in coupled runs); this was required to be able to use the commits from develop: merge HWRF version of saSAS with GFS version ufs-community/ufs-weather-model#94 directly so that once this PR is merged into develop, it can be brought back to NCAR dtc/hwrf-physics with a clean commit history (all commits already included) and no merge conflicts

Associated PRs:
NCAR/ccpp-physics#433
NCAR/fv3atm#41
#39

For regression testing information, see below.

[climbfuji]

Regression testing on hera.intel:

First, run against existing baseline. The following tests fail (all expected):

fv3_stretched_nest (PROD)
fv3_stretched_nest_quilt (PROD)
fv3_regional_c768 (PROD)
fv3_ccpp_gfdlmp_hwrfsas (REPRO)
fv3_ccpp_gfdlmp_hwrfsas (PROD)
fv3_ccpp_stretched_nest (PROD)
fv3_ccpp_regional_c768 (PROD)

Out of those, all tests that contain stretched_nest and regional_c768 fail in PROD mode, because the blocksize in input.nml is changed as part of this commit. This is to avoid non-uniform block sizes, which creates remainder/peel loops when the code is vectorized and often leads to b4b differences with the Intel compiler when the AVX2 SIMD instructions are used (i.e. in PROD mode).

The remaining two tests fv3_ccpp_gfdlmp_hwrfsas in REPRO and PROD mode are new and thus have no baseline to compare against. A manual inspection shows that they run to completion.

rt_hera_intel_against_20200317_baseline.log
rt_hera_intel_against_20200317_baseline_fail_test.log

Second, create a new baseline for these failing tests only. All successful. Sync old/existing baseline /scratch1/BMC/gmtb/ufs-weather-model/RT/NEMSfv3gfs/hwrf-physics-20200317/ to new baseline /scratch1/BMC/gmtb/ufs-weather-model/RT/NEMSfv3gfs/hwrf-physics-20200410/, then sync the newly created baseline results for these tests into it.

rt_create_new_baseline.log

Third, run full set of regression tests against this new baseline. Two tests failed due to super annoying, reoccurring filesystem issues on hera in the first round:

fv3_ccpp_thompson
fv3_ccpp_wrtGauss_nemsio_c768

rt_hera_intel_against_20200410_baseline.log
rt_hera_intel_against_20200410_baseline_fail_test.log

Rerunning those tests only: all pass.

rt_hera_intel_against_20200410_baseline_rerun.log

[climbfuji]

These PRs are ready to merge.

[grantfirl]

Looks good to me.