Learning with network arrays

[studywolf]

made a branch to attack this, it could be a bit of a journey... https://github.com/ctn-waterloo/nef-py/tree/na_learn

currently test_learn_array runs in the case where pre is guaranteed to be 1D and post / error signal are ND

update:

• now runs with ND pre / post / and error, just need to submit it to extensive testing and it'll be good to merge back in!