Introduce merge-tree abstraction

[vasia]

For the bipartite detection algorithm and the top-k degrees.

[balidani]

So far we have two implementations that use a merge-tree and here is what it looks like when we create one:

edges
        .map(new InitCandidateMapper())
        .map(new BipartitenessMapper())
        .groupBy(new MergeTreeKeySelector(0))
        .map(new BipartitenessMapper())
        .groupBy(new MergeTreeKeySelector(1))
        // Repeat the above 2 lines log(DoP) times
        .map(new BipartitenessMapper());

The initial mapper just converts the edge to a format that is used by the main one. What do you think, what would be a good way to express this in gelly-streaming?

edges
        .mergeTree(new InitialMapper())
        .with(new SomeMergeTreeMapper())
        .on(0);

I was thinking of something like this, probably with different wording for the function names. The last part of it just receives the position of the key to later group by to form the tree.

[balidani]

This is implemented now