Implement Dynamic Splitter for Train and Test Datasets

[unibeck]

Currently we split the dataset into an industry standard of 70% for training and 30% for testing. However, we simply grab the first 70% of the original dataset for training and the last 30% for testing.

This does not lend to the best training and determining the model's accuracy. Instead, we should sparsely split the data over the entire dataset evenly.

https://gluon-ts.mxnet.io/_modules/gluonts/dataset/split/splitter.html https://gluon-ts.mxnet.io/api/gluonts/gluonts.dataset.split.html

[unibeck]

This was completed within #13 and #14