Derived variables structure update

[justin-richling]

Move the calculation of derived variables out of adf_diag.py into an external script. This is a placeholder until a more robust variable derivations scheme is implemented.

This will help declutter adf_diag.py in the hopes of keeping non-essential methods separate.

adf_derive.py has two main methods: check_derive and derive_variable

• check_derive - called in variable loop; will check on each available variable
  For incoming variable, look for list of constituents if available as a list in variable defaults yaml file

  If the variable does not have the argument derivable_from or derivable_from_cam_chem,
  then it will be assumed not to be a derivable variable, just missing from history file

  If the variable does have the argument derivable_from or derivable_from_cam_chem,
  first check cam-chem, then regular cam and add all constituents to diag_var_list for time series generation.

  NOTE: this will only modify diag_var_list for use in adf_diag.py, it will not modify the yaml variable list used in other scripts.

• derive_variable - called after variable loop and time series generation; loop over dictionary containing derived variables as keys and list of constituents as values.
  This will then be called in a loop of accepted derived variables and will be generated from constituent time series files.

[nusbaume]

Nice cleanup! Just have a few suggestions.

[nusbaume]

Instead of checking whether constit_list is a list type below, would it possibly be safer to just add a logical that skips writing the constit_errmsg if you have already written it here?

[nusbaume]

If you use the logical method suggested above, then you can probably get rid of the isinstance check here.