Add an end-to-end quality test

[eu9ene]

Due to the recent bugs in inference, it would be nice to have a unit test that runs translation on a test set end-to-end using a browser and checks translation quality.

For example, we could,

1. Take our regular flore-test test set for one or several language pairs
2. Create a web page where each line in the dataset would be a separate paragraph
3. Run translation in the browser
4. Take the output page and split it back
5. Calculate the COMET or BLEU score
6. Compare the score to the evaluation results. It should be similar with a small error.