fix decomp test failure by broadcasting calendar fields

[eclare108213]

Fix nonBFB decomposition test comparisons.

• Developer(s): E. Hunke

• Please suggest code Pull Request reviewers in the column at right.

• Are the code changes bit for bit, different at roundoff level, or more substantial?

BFB except for the tests that were failing.

• Is the documentation being updated with this PR? (Y/N) N
  If not, does the documentation need to be updated separately at a later time? (Y/N) N
  Note: "Documentation" includes information on the wiki and .rst files in doc/source/,
  which are used to create the online technical docs at https://readthedocs.org/projects/cice-consortium-cice/.

• Other Relevant Details:

This PR addresses issue #137. All of the ranks/tasks need calendar information for the frz_onset and mlt_onset variables, but nyr was not broadcast when using restart calendar information. This broadcast has been added for all three I/O types (only netcdf is tested here). We might also need to broadcast month, mday and sec too, but I’ve elected to wait on those until we actually see a problem with them.

I also updated drivers/cice/CICE_RunMod.F90_debug to match the current module. This allows helpful printing of many pertinent fields throughout the timestepping for a grid point specified in ice_diagnostics.F90 (I added frz_onset to the printing routine temporarily but am not committing that change here).

I have tested base_suite and decomp_suite, but using pes=1x8x5x29x80 in decomp_suite instead of 1x20x5x29x80 (see issue #137). The boxrestart test fails (known issue, #138) in base_suite, otherwise everything passes.

[apcraig]

Are the changes in CICE_RunMod.F90_debug just to bring it up to date? I hadn't really tested that, is it working now?

[eclare108213]

Yes, this just brings CICE_RunMod.F90_debug up to date. If you diff that with CICE_RunMod.F90, the only differences are the lines needed for printing out diagnostic info.

[eclare108213]

And I used it to track down this bug, so consider it working...

[dabail10]

Is this also a problem in CICE5? I noticed in our version that the io_pio version was reading in nyr on all tasks. Here the broadcast is not necessary. Our io_netcdf and io_binary have this issue, but we never run with these in the CESM.

[eclare108213]

I'm not sure whether we had the namelist option use_restart_time in CICE5, but even if we didn't, this might still have been a problem. It's rare though, not something I would have noticed even if I were paying very close attention to frz_onset and mlt_onset, which are just kludgey diagnostics.