Extend varinfo to include other normalized variants.

[jpfairbanks]

Variation of information can be normalized in different ways based on the properties of the application.
I added a symbol argument to specify this normalization to the varinfo function and set the default to the previous behavior. I also put the computation of I, H1, H2 into a function information so that it could be reused in the future. This is a nonbreaking change.

If there is a better way to do it, let me know and I can change the pull request. I think this is the least disruptive change possible. I added tests and documentation to match what was there.

The paper cited is here
http://jmlr.csail.mit.edu/papers/volume11/vinh10a/vinh10a.pdf

[pcjentsch]

This PR would be useful to me, what prevented it from being merged?

The same authors that @jpfairbanks cites have since published a follow-up paper, which brought me to this PR.

[alyst]

@pcjentsch I guess it was a lack of maintainers resources :( Unfortunately, the PR was submitted quite long time ag, and the codebase has changed since. To the level that there are merge conflicts, and it cannot be merged automatically. If @jpfairbanks would be so kind to review and rebase his work, I would be happy to review it. Alternatively ‐ if he would be ok with that - you can fork this PR and submit an updated version.