Train SentenceBERT model

[jordiae]

Use https://github.com/UKPLab/sentence-transformers

It's very easy to use. Use it as feature extractor for the moment, attach a simple linear classifier for the moment.

[asier-gutierrez]

Pending

• Add seed for everything. Additional isolated one for dataset split indexes based on https://stackoverflow.com/questions/49555991/can-i-create-a-local-numpy-random-seed.
• Check that validation and test works. Train seems to work.
• Verify the complete functioning of the process.
• @jordiae Check batch/sub-batch mismatch during training.