Add more notebook datasets, test depth=16

[RAMitchell]

Implements #38

[mfoerste4]

It is a bit hard to review these changes as the file is stored in git LFS. I have browsed over it and I have a couple of thoughts:

• This notebook has grown quite large. Would it be feasible to extract the utilities (data fetchers / runners) into a python module?
• I don't quite like the inclusions of graphs & performance numbers into the file that is checked in. These numbers are not auto-updated once we change the code and they are only valid for a certain hardware they were run on the last time someone modified the file. I would also prefer to not have the verbosity log all iterations into the notebook output. Maybe we should split it into a (small) notebook example with graphs and a larger benchmark comparison script (that can also output to csv?).

[RAMitchell]

I'd rather not extract them at the moment but keep it a self contained example. The notebook is not that large as loading each dataset is only a few lines.

I guess thats the problem with notebooks in general, but I still think its worth checking in even if it might go out of date. Agree with removing the verbosity output.