Pushing update for metacat training

[shubham-s-agarwal]

Revamped the training for both basic and advanced notebooks with focus on simplifying the training, especially for 2-phase learning.

[mart-r]

The changes to meta_annotation_training.ipynb look good. Just simplifying things since you can load both LSTM and BERT models using the API rather than doing the different steps manually. Really good change IMHO.

With respect to meta_annotation_training_advanced.ipynb:
There's clearly some simplification just like in the other file. And that's great!
And there's some simplification when you get to the 2-phase learning as well. Though perhaps - to keep in line with the previous version, you could set the per-phase class weights in this part? Instead of providing the empty lists, you could use what was used before (and is provided in the demo after).
And then I'm not sure about the demo part. Looks like that's new (i.e in addition to what was included before). What value does the demo provide? It seems like the code at the end of the previous section already performs the 2-phase training of the defined meta cats. And the "demo" code wouldn't even really run since the method signatures aren't being followed. Yet there's output to it, which would suggest that it could be run, even though the disclaimer above says not to. I don't know, just feels a little confusing.

[shubham-s-agarwal]

I have removed the demo cells to make it easier to follow

[mart-r]

Looks good to me