Hello,
I have a program supporting multiple datatypes of .npy files. Currently, I think the only way to handle datatypes not know at compile time with ndarray-npy is to handle errors of read_npy.
One solution could be to expose header parsing (as in #81) to get access to the descr field.
