Improve translation of URLs

[eu9ene]

Sometimes URLs are written in text rather than hidden behind the HTML element. The URL should be copied as is in this case.

There are two ways to fix this:

1. Maybe an easier way: identify a URL with a regex on the translation engine side and copy it without passing to the model
2. Add data augmentation to insert URLs in some training examples and retrain the models

[marco-c]

Could also be a data cleaning problem, like num_mismatch.

[gregtatum]

I would push back against implementing option 1, which would happen on the Gecko side for every translation. That regex seems risky and error prone to write. I would at least start with data augmentation. There is https://github.com/hplt-project/OpusTrainer/issues/43 already on file.

[eu9ene]

See also https://github.com/hplt-project/OpusTrainer/issues/43