Add python module load command for izumi

[peverwhee]

To ensure CAM runs with python 3.7.0, add a module load command to the izumi section in config_machines.xml.

< command name="load" >lang/python/3.7.0

[rljacob]

You're supposed to load your python environment prior to running any cime commands. Doing it within cime leads to chaos.

[billsacks]

Under some circumstances, you need to do a module load python within cime in addition to having the right python module loaded in your own environment. e.g., I needed to add one for cheyenne for scripts_regression_tests to work correctly (#4033). I think the problem in that situation was that the overall python process (scripts_regression_tests) was spawning other CIME python processes (e.g., create_newcase), some of which were resetting the module environment according to config_machines, and this included a module reset that cleared any personally-loaded python module.

I can envision other possible issues that might occur even within a single create_newcase/case.setup/case.build/case.submit workflow: CIME calls things like buildlib as a python subprocess, so these scripts might be executed with the wrong python environment under some circumstances.

To deal with issues like this and other possibly unforeseen issues, I would support having an explicit module load of the correct python module. I agree with @rljacob that this mechanism can lead to chaos, though, and long-term we really need to think about #4059 .

[gold2718]

This is a CESM-only change (where there are already 11 explicit python module loads including 10 which appear to be loading a python 2.7 version).

[gold2718]

There has been a discussion over in #4117 about whether and why python 3.7 might be needed in CAM.
The new answer (after much testing) is that we do not need python 3.7 to build and run CAM.
However, we would like to use at least one python3.7 feature to clean up our code and for running tests.
An example of the change is in the "type" of a compiler regular expression.

In python 3.6.8 (default on Izumi):

>>> import re
>>> foo = re.compile(r"hi mom")
>>> type(foo)
<class '_sre.SRE_Pattern'>
>>> isinstance(foo, _sre.SRE_Pattern)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
NameError: name '_sre' is not defined
>>> bar = foo.match("hi mom, how are you?")
>>> type(bar)
<class '_sre.SRE_Match'>

In python 3.7.0:

>>> import re
>>> foo = re.compile(r"hi mom")
>>> type(foo)
<class 're.Pattern'>
>>> isinstance(foo, re.Pattern)
True
>>> type(bar)
<class 're.Match'>
>>> isinstance(bar, re.Match)
True

So both in code (wanting to use duck typing to catch an input that may be a regular expression pattern) and in doctests (e.g., did a routine correctly return a regular expression match), python 3.7 offers a cleaner approach.