Operate under Inference Mode as well as No Grad

[jatkinson1000]

Originally commented by @ElliottKasoar as part of #103

• InferenceMode
  No changes are currently included, but it would be good to support InferenceMode too eventually, as it should provide further performance benefits over NoGradMode.
  However, it has stricter requirements, and the mode was only added (as a beta) in PyTorch 1.9, so we would need to be careful if we want to support older versions.

"Looking at the docs we'd maybe want to create an optional argument bool :: inference mode and if set to true enable c10::InferenceMode for the function as a block.

[jatkinson1000]

Copying @ElliottKasoar's comment from #81

2. InferenceMode

• See: inference mode, autograd mechanics and the dev podcast
• From our benchmarking (see FTorch with InferenceMode and NoGradMode sections), benefits were less clear, but in general it is expected to be at least as fast
  • Tests were carried out by replacing torch::AutoGradMode enable_grad(requires_grad); with c10::InferenceMode guard(requires_grad); in all ctorch.cpp functions, but ideally both options would be presented to users
• This mode was only added (as a beta) in PyTorch 1.9, so we would need to consider support for older versions
• The mode is also much stricter than NoGradMode, so cannot be used in all cases

[jatkinson1000]

@jwallwork23 and I just discussed this.

• Because we do not have a single C++ program and scope we can't simply set the value once and would need to do it function-by-function.
• Providing it as an optional argument to every function is not ideal as it presents the opportunity for user error (passing one value to tensor creation, but a different one to net etc.)
• The ideal solution would perhaps be to have a boolean in FTorch that is used to set the value of InferenceMode when calling to C++ functions. We would provide a function set_InferenceMode() or similar to set the value of this for a 'block' of Fortran to emulate a scope.

This would be good for a hackathon.