Add support for parallel-text

[mrunesson]

We need to support packaging of parallel texts.

This means merge the two fields into text in groups of N(40?)
Result will be something like: src_lang: xxxxxx\ntgt_lang: yyyyyy

See discussion in https://mattermost.ufal.mff.cuni.cz/openeurollm/pl/xj3y5zfxmiy7mr4ftw15aqreuw

[mrunesson]

#36 add support to do the merge of parallel texts. It is not yet integrated into pipeline.

[mrunesson]

Sample so a document either has lang1-lang2 or lang2-lang1. It shall be 50-50 over dataset. https://mattermost.ufal.mff.cuni.cz/openeurollm/pl/nw181g7737b3xr644r7p9q671a

[mrunesson]

Left is to verify it works with contamination data. PII will not be included according to OpenEuroLLM/training-data-collection@c52253c#diff-23e63c476a385b3bebe43e0100bf2caaeaac3f725ffae5b748793375ff068e8e