Add optimisers

[jatkinson1000]

This PR seeks to add optimisers functionality to the code.

• FTorch
  • Expose optimisers in FTorch
  • Write unit tests to cover these
• Exercise:
  • README - WIP
  • requirements
  • python version of the exercise
  • Fortran version of the exercise - WIP

Notes whilst in progress:

• It may be useful to expose a few more tensor functions such as .mean() and .sum() - see Overload sum intrinsic for tensors #240
  • it is handy to implicitly initialise a gradient of ones in .backward() by contracting a (loss) tensor into a scalar. Should this be added to the autograd example?
• ftorch.F90 is going to grow with this. Perhaps now is the time to break apart into sub-modules?
  • I'm thinking of having ftorch_optim as a module to hold these.

[github-actions]

Cpp-Linter Report ⚠️

Some files did not pass the configured checks!

clang-format (v12.0.0) reports: 2 file(s) not formatted

• src/ctorch.cpp
• src/ctorch.h

Have any feedback or feature suggestions? Share it here.

[jwallwork23]

At some point it might make sense to separate out modules for tensor, model, optim.

[jatkinson1000]

Yeah, since each optimiser (Adam, SGD, etc.) seems to be its own function I thought I'd do this here, for the Fortran at least.
I had been hoping that there was a general optimiser function specified by an enum, but alas no.

Others feel OK for now, but definitely something I was thinking about, especially as module functions might grow soon.