Split by lines

[xpert13]

Add support for sentence splitting. Currently text is sent to the LLM as-is, but longer texts (like documents) should be split into chunks, translated and merged back.

I propose to do what DeepL (or even Google Translate) does - break it down into lines. They translate each line separately, and if you send 20 lines for translation, you will always have 20 lines as a result.

[christopherpickering]

Maybe that will help speed it up as well: I was trying 3000-4000 char paragraphs on a rtx3090 and responses were almost a minute w/ the 27b, and the 12b was only a bit faster, but doing sentences is pretty quick.

On the 3090, it was only going to around 40% utilization when doing 4000 char paragraphs, so I don't think it was a bottleneck.