PyTorch GWP model for MiMA v2.0

[dorchard]

• Make it
• Refactor it a bit so it fits better (in terms of input/output tensors) and only uses the variables it needs.

[jatkinson1000]

Tackling this in the pytorch branch

Tasks:

• Obtain PT wavenet
• export PT model to TorchScript
• Remove calls to existing GWP in MiMA (from atmos_param/cg_drag/cg_drag.f90)
• Bring relevant variables for NN to atmos_param/cg_drag/cg_drag.f90
• Convert inputs to tensors relevant for PT
• Call TorchScript
• Update model namelist/initialisation procenures
• Build process/ compile
• Test
• Document

Following this a branch should be made with the PT parts removed to tackle #5 ready for Simon to tackle in #3

[jatkinson1000]

This is now written in the pytorch branch of DataWave MiMA.
There is also an equivalent binding using forpy in the forpy_davenet branch of DataWave MiMA

Both of these compile on csd3 but are yet to be tested.