Generate data for each stage of the ML process

[jatkinson1000]

In order to compare and test the coupling routines we need data from the Python to compare to.

This needs:

• Input data for each variable input to the net
• Normalised variables (following the normalisation code) as provided to the net
• Normalised tendencies as output by the net
• Denormalised tendencies as used in CAM

Running the python code to generate these and place them in a NetCDF will allow a comparison with the Fortran to be made.

[jatkinson1000]

The NaN issue is resolved in #14 so this is no longer critical.
We can revisit this oin future if we feel it would be of benefit, but for now resources are better deployed elsewhere.