This first commit is done in anticipation of supporting time series files in
addition to history files, and also using intake-esm catalogs behind the
scenes.

1. start_date & end_date are not part of the class constructor
   -- basically all the __init__ routine should do is set up a catalog
   -- _find_hist_files() also doesn't rely on these variables either; this
      routine is essentially building the catalog, and we don't restrict the
      files listed until we are ready to call open_mfdataset()
2. _open_history_files() -> gen_dataset(); this routine is where start_date and
   end_date are specified, and it returns a dataset rather than updating a
   class member variable. Logic to apply start_date and end_date at this stage
   is much simpler than doing it in _find_hist_files() (or I figured out a much
   easier way to do it). This routine also now expects a list of variable names
   to include in the dataset.
3. compare_fields_at_lat_lon expects array of DataArrays, not array of cases
4. Added get_varnames_from_metadata_list() to utils/utils.py because I need
   this function to generate the list of variable names for gen_dataset() in
   several notebooks.

I also created a script that submits jobs to the slurm queue to re-run
notebooks, though it does not work for the trend_maps or plot_suite_00[34]
notebooks (I think because of the reliance on dask and NCAR_jobqueue).

gen_dataset() first reads time series from campaign, then reads any additional
data from history files (from archive or run directory). There's an API change
where, instead of expecting start_date and end_date (strings formatted as
'YYYY-MM'), the function wants integers start_year and end_year. Figuring out
better default values will be necessary to extend this to other CESM runs but
it's a good start.

I split plot_suite_004.ipynb into 4-year segments, as it was choking on 8 years
of output. I also added plot_suite_map notebooks for 004 for years 0006 and
0007 (we're one month shy of 0008).

I had left the "import yaml" statement in from when I toyed with the idea of
passing the metadata yaml file to __init__

Concatenating time series dataset with history file dataset now works as
expected

Also reran plot_suite notebooks in an environment that will match previous
master (note that this changes plots in Sanity Check). Next commit will be
updated trend_maps, and the plan is to squash-merge this PR to remove the
commits that changed the plots and then changed them back.

Re-ran in an environment that matches the previous master to minimize diffs in
the PR.

I didn't realize this notebook hadn't been run in previous commits

Some of these were plotting the wrong year; I refactored all of them to be more
clear about what is being plotted.

For now the comparisons are done based on variables in diag_metadata.yaml.
Required a flag to suppress some CaseClass output.

All plot_suite_maps now match master

* run_notebook.sh launches a single notebook via slurm, and run_all.sh is now
  run_all.py (python-script) that calls run_notebook.sh
* _find_timeseries_files() now includes pop.h.nyear1 files
* gen_dataset():
  1. no longer assume time_bound is the name of the variable (look at bounds
     property of ds["time"]
  2. if bounds are not decoded, still update start_year before looking for
     history files (via decoding time, which may not be best solution)
* use list.extend() instead of += operator