Generalize Final data.npz output

Currently the `data.npz` output, by default is split into a `test` and `train` set with 80% of the data going into training and 20% of the data going into the test set. While the ratio, number of split sets, and names of split sets can be changed, it could be better to return a `data.npz` and let users handle splitting the data outside of label-maker for increased flexibility.

I think it could be useful if the new version of the `data.npz` file returned as the `x-y-z` tile id as the key and the values is a list with 2 elements, the first being the label in numpy array format, and the second being the image in numpy array format.

Another benefit of this change will be that the image label pairs in numpy array format will retrain the `x-y-z` tile id. The `x-y-z` tile information is currently only associated with the `labels.npz` file and not associated with the `data.npz` file.

thoughts? @drewbo