exp: adds MLP and LR with anomalous data in the training dataset

[lanterno]

closes https://github.com/c4dt/predictive-maintenance/issues/11

This PR adds two new notebooks that use anomalous data as part of the training dataset for the transfer mode (MLP and LR).

Using anomalous data certainly shows better results than without.

MLP:

Train RMSE: 0.89
Train R^2 score: 0.95
------ 
Test RMSE: 2.74
Test R^2 score: 0.66

LR:

Train RMSE: 1.88
Train R^2:  0.76
------ 
Test RMSE: 3.03
Test R^2:  0.582

We can also see here that the image anomaly score is very close to patchCore

MLP:

LR: