Add a parameter to load a model for the conversion method

[TheJoin95]

At the moment we're shipping the model files within the package which I believe it's incorrect: it's adding a lot of MB in a python package which it's working even without it.

I believe we should add some parameters to load the models. Eventually, we should provide a way to add a custom model by declaring the input shape/size and the network.

These could be split in two issues:

1. add parameter to load the PaletteNet model
2. add a custom model, customisable