GMM covariance types examples overly complex / confusing

http://scikit-learn.org/dev/auto_examples/mixture/plot_gmm_covariances.html#sphx-glr-auto-examples-mixture-plot-gmm-covariances-py

I don't particularly like the example because it's supervised and uses a train/test split and identification with the original classes. I think it would be better to use a synthetic dataset and just show off the different covariance types.

It also fits the model on 4d data and only shows a 2d projection and that's not super intuitive imho.

Also, the example could be much simplified if we added a "get_covariance" function back to the model. I think we had that in the old GMM. Was there a reason not to add it to the new GMM? In many cases the user wants to be agnostic to the storage format of the covariance matrix, I think.

Start with the linked auto_examples/mixture/plot_gmm_covariances.py example and review how the current GMM model stores covariance data. Run the example to understand its supervised train/test flow and 4D-to-2D presentation. Done means the example clearly compares covariance types on a synthetic dataset, with the proposed get_covariance API considered and its behavior documented or tested.