Starting graph algebra

[Mec-iS]

First PR for #261

Please suggest how to proceed. For now I have implemented conversions to numpy.array, which metrics/algorithms would you like to see computed?

Please consider writing notebooks that use these functionalities.

cc: @tomaarsen @ceteri @SultanOrazbayev

[ceteri]

Looks good!

[ceteri]

This is looking good!
Two questions:

1. Dependencies numpy, sklearn, skn are pegged to specific point releases. Do they need to be? Previously, this caused lots of compatibility problems since people on different platforms (e.g., Windows) working on different specific use cases (other dependencies) ran into many conflicts, since they required specific point releases. Is there a way to use ranges instead?
2. For network analysis, is there any trade-off comparison of using these approaches (e.g., with skn) versus using NetworkX? If we use NetworkX then we can often pickup paths for better performance using GPUs (RAPIDS) and scale.

[Mec-iS]

1. These are the current dependencies situation: scikit-learn has to be pinned to current stable version 1 as when the current nightly version 2 will come out there may be breaking changes, skn requires numpy==1.23.0. For Windows users, may be useful to suggest using Anaconda or implement the usage of poetry that I was thinking about implementing anyway.
2. Yes you are right. I thought to use skn as the API is very user-friendly and similar to the very popular scikit-learn, in terms of GPU support networkx is probably better.

I will open issues for some of the above, let me know which ones.

[Mec-iS]

@ceteri #271