MLP on permutedMNIST

[karangrewal]

Here, I do the following:

1. add configuration files for training a 3- and 10-layer MLP on permutedMNIST, and
2. add an MLP class with modified weight initialization used by the configs to nupic/research/frameworks/pytorch/models so future MLP experiments can also use these.

[karangrewal]

@benja-matic This is good feedback. To address your concerns,

1. I moved the MLP class to nupic/research/frameworks/pytorch/models and renamed it ModifiedInitStandardMLP and included some documentation there, and
2. I added comments regarding hyperparameter choices so they don't seem arbitrary.