Kmeans training step should check for numpy arrays

[k034b363]

Is your feature request related to a problem? Please describe.
Currently, train.kmeans does not check the extensions of files in the input directory, so if intermediate or hidden files are added to the directory during workflow development, they sometimes cause problems.

Describe the solution you'd like
A check for each file that whatever gets read in results in a numpy array.

Describe alternatives you've considered
An alternative is to check the file extension of each file in the directory, but that seems like it would be a more complicated check than just checking if something is an array.