Add parameter to select model name

[Kshitij68]

#423
This is introductory PR which introduces the basic strucutrue of the extension that I propose. I think further testing can be done as well.

[pplonski]

Thanks @Kshitij68, I would leave _best_model in the base_automl.py - the information about best model will be needed anyway, we shouldn't loose it.

The basic structure of the extension that you propose looks good. I'm very interested in the full structure of the extension. Can't wait for it!

[Kshitij68]

I'm trying to add more test cases.
The only way I see to assert that the correct model is used is via asserting the log loss or accuracy and comparing with the values that are persisted.
However, this approach doesn't work as the log_loss doesn't match between training and prediction dataset?