Clean up RNN branch about grainlearning

[APJansen]

With poetry we probably want to make the rnn dependencies optional, as they are quite heavy (mainly tensorflow).

[APJansen]

Another todo: generate the documentation, and probably for that to work we have to pick one convention for docstrings as currently we're using different ones.

[APJansen]

I manually updated pyproject.toml adding the rnn dependencies. For the versions I checked what the latest was and used that, but that might not be necessary. Then ran poetry lock --no-update to add these to the lock file without upgrading versions of other packages. Then tested installing with poetry install, but there is an issue with tensorflow, specifically: Unable to find installation candidates for tensorflow-io-gcs-filesystem (0.27.0) (related issue)
Likely this is because I'm on a (arm) mac, replacing tensorflow with tensorflow-macos fixed it for me, everything now installs. But this needs to be fixed for everyone somehow..

[APJansen]

It runs now, as long as the data is present at grainlearning/rnn/data/sequences.hdf5. Wandb creates a folder inside rnn/.

[luisaforozco]

We picked the following convention for docstrings:

• Use python type hints.
• Use sphinx format for docstrings, specifically:
  :param : and :return :, we are dropping :type : and :rtype:

[luisaforozco]

Regarding the dependencies, In this commit I created two groups of extras rnn and rnn_M1_macOS with the necessary dependencies (only difference is in tensorflow).
I tested in my machine poetry install -E rnn_M1_macOS and on Snellius poetry install -E rnn, and the dependencies are handled correctly.

[luisaforozco]

The installation of tensorflow for arm64 macOS requires some extra steps that cannot be handled from poetry. I think the best is to keep it separately: if an user has M1 macOS he/she will have to install tensorflow first on his own, and then he/she can install grainLearning on the same python environment.