Add general weighted matching problem

[simonbowly]

Why this Nup?

Maximum (weighted) matching. This is a distinctly different model (MIP) from the linear network flow of bipartite matching.

Does it fall under an existing category?

Yes, graphs

What will the API be?

scipy sparse graph input, scipy sparse output (max. matching as a subgraph)

Additional context

We want to have fit-for-purpose models where we assume:

• Users would know graph theory well, so would be aware their graph is bipartite
• They would not want to see a MIP formulation when they know a fast LP algorithm is more suitable

[simonbowly]

I've removed general weighted matching for now with b5855e2. Will open a PR (starting with reverting this commit) if we want to restore it at a later date.