Random-neural-net-models

fastai based `Learner` with callbacks enabling
* unified training and inference with `tensordict`
* tracking of activations and weights
* learning rate search
* hyperparamter scheduling (e.g. one cycle)

tabular models
* supervised (classification / regression) for numerical & categorical features, including handling of missing values
* unsupervised (variational auto encoder) also for numerical & categorical features, including handling of missing values

Added mingpt (source: https://github.com/karpathy/minGPT/tree/master), including the three projects (adder, sort, char) with dedicated notebooks.

added Diffuser UNet based on https://github.com/fastai/course22p2/blob/master/nbs/26_diffusion_unet.ipynb

refactored unet implementation in `unet.py` to improve readability

* new `unet.py` contains modified version of fastai 2022 unet implementation
* added `unet_fastai2022.ipynb` to apply implemented unet to mnist

* added fastai 2022 course like resnet implementation in resnet.py
* applied resnet to mnist in resnet_fastai2022.ipynb