Corpora exclusion rules

[ZJaume]

The skip_datasets in the config could be improved, I think.

• The Multi* versions should not be excluded, still valuable. Specially if the language pair is not english-centric.
• If SPC is not failing anymore, should be removed. For the two or three language pairs that I have taken a look in this corpus, it was quite clean. Mightbe a good resource.
• There should definitely be another skip_datasets by language pair. For example Ubuntu and PHP are full of garbage for Chinese but are good for other language pairs.