Add preprocessing factors to json file

[scarlehoff]

Since I was already looking at what was (or not) independent of the length of the basis I decided to also add the preprocessing factors to the output json file, which I believe was a feature of nnfit that was lost in n3fit just out of laziness.

Example output:

  "preprocessing": [                                                                                                                                                                           
    {                                                                                                                                                                                          
      "fl": "sng",                                                                                                                                                                             
      "smallx": 0.4398651421070099,                                                                                                                                                            
      "largex": 2.6411304473876953,                                                                                                                                                            
      "trainable": false
    },
    {
      "fl": "g",
      "smallx": 0.734717071056366,
      "largex": 1.709303855895996,
      "trainable": false
    },
    {
      "fl": "v",
      "smallx": 0.10663021355867386,
      "largex": 2.4233202934265137,
      "trainable": false
    },
    {
      "fl": "v3",
      "smallx": 0.7327896356582642,
      "largex": 2.9963228702545166,
      "trainable": true
    },
    {
      "fl": "v8",
      "smallx": 0.6589133739471436,
      "largex": 3.629549741744995,
      "trainable": true
    },
    {
      "fl": "t3",
      "smallx": 1.161284327507019,
      "largex": 1.8856220245361328,
      "trainable": true
    },
    {
      "fl": "t8",
      "smallx": 1.1631083488464355,
      "largex": 4.324087142944336,
      "trainable": true
    },
    {
      "fl": "cp",
      "smallx": 0.5860853791236877,
      "largex": 2.698132038116455,
      "trainable": true
    }
  ],

[scarlehoff]

Updated with the comments form the code meeting.

[scarlehoff]

I've added a get_weight_by_name method to the layers so that one can get the right alpha and beta without having to care about the order. If there's a mismatch one would see null in the json file, which I guess is better than just skipping the dictionary.

[scarlehoff]

Forgot to look at the calendar, is "Travis is broken for some reason" day again! :)

[siranipour]

The new commits still look good to me.

The failing tests are because of this apparently

[scarlehoff]

Funny it never failed before. I'll push a possible fix.