Pass "explaination" of each string as context to LLM / OpenAI translators

[acgourley]

At my company we have noticed GPT-4o misses the mark on translations for buttons or short titles, and believe the main issue is the model not having enough context. If we could inform the model that a string will go on a purchase button or section title surely it will perform better.

The simplest solution I can see is to expand the prompt to not just add the texts but to include the weblate explanation.

In the case of OpenAI, moving to their structured outputs API might be good to do as well, it should simplify openai.py and reduce token usage.

I'm happy to make a feature request but I wasn't sure if that's the correct next move as I'm not active in this github project.

[nijel]

What we would also need is structured input. We currently pass several strings at once to LLM to improve performance and cost effectivity, but passing additional per-string data is problematic with this. See also #13398