Create notebook with exercises to fill in

[jopasserat]

Inspired from Flower's tutorial: https://colab.research.google.com/github/adap/flower/blob/main/tutorials/Flower-1-Intro-to-FL-PyTorch.ipynb#scrollTo=J4Em7BPNTXeX

Should run successfully in a Colab environment

[jopasserat]

tentative outline

1. simple FedAVg with IID data
2. same FedAvg but non-IID split (adjust the allpha value in flower's dirichlet's implementation)
3. play with number of local iterations to highlight impact of non-IID / client drift
4. test other agg strategies to address the non-IIDness
5. introduce DP-SGD via Opacus using this other example https://github.com/adap/flower/blob/main/examples/opacus/dp_cifar_simulation.py