add predict method for new data & test batch/streaming using method

[JesseKolb]

The TICC object will save its data in self.trained_model because it seemed the current output of fit(), train_cluster_inverse , is not enough by itself to predict for new data. Once fit, the predict_clusters method can be used to predict clusters on new data by passing it in as an argument.

I just added 2 basic tests using this as examples, but I'm not sure if it'd be best to refactor the unit tests, since I just added it to one of your tests. I think it'd be good to add more unit tests as well going forward, as well as perhaps some short docstrings to some of the functions which are hard to understand without context (e.g. upper2Full and Prox_logdet)

[davidhallac]

The predict_clusters method looks great - thanks!

The unit tests look good, though yes, we are hoping to revamp our unit tests in the near future to be a lot more comprehensive/robust (we initially just added small unit tests to integrate a few new changes we were making to the code, but as more people use TICC, I agree that we need to refactor and continue to improve them)

[heastdream]

It is hard to understand the function of upper2Full and Prox_logdet in admm_solver.py, can you give me some help