chore(#2199): update steps on how to add new keys and LLM use

[mrjones-plip]

Description

Update the steps on how to add add new keys to reduce requirement on having native speakers supply translations.

closes #2199

License

The software is provided under AGPL-3.0. Contributions to this project are accepted under the same license.

[mrjones-plip]

@andrablaj - I made the text intentionally a bit less prescriptive than we were discussing in #2199. My thinking is that the steps guide folks in the right direction and ultimately we have to trust they do the right thing. During the stewardship call folks explicitly mentioned how LLMs were better than services like Google Translate, so I left that suggestion out.

[mrjones-plip]

cc @dianabarsan or @jkuester in case you want to have a look-see as we talked about this on the stewardship call.

[andrablaj]

Thanks for turning this around so fast! 🚀 A few suggestions:

1. The current text calls out LLMs specifically, but someone using Google Translate, Transifex, or a bilingual friend would be unsure if the same rules apply. Suggest: "you may use translation tools (LLMs, machine translation, etc.) or ask a fluent speaker."
2. Given CHT runs in real health systems, I think a do-no-harm lens has to win — if the contributor isn't fluent, we should require community validation before merge, not after. So the trigger becomes "any translation you didn't author as a fluent speaker" rather than "any LLM translation."

Happy to discuss further

[mrjones-plip]

Looping back in this Slack thread where @jkuester said::

you are correctly presenting our original conclusion from the Dev Hour(?) call (where Andra was not present 😅).  One thing we discussed was if a rough-or-even-wrong translation was better or worse than no translation (falling back to English or just the key). Several people noted a preference for a rough-or-even-wrong translation while no one was enthusiastic about no translation.

As I mentioned during that call, as someone who barely speaks one language I do not feel qualified to speak to the safety/correctness of AI translations. I am happy to follow any process here that we determine is best. 👍

@andrablaj has extensive language knowledge as well as more experience with AI translations than anyone else I have talked to. So, honestly, if she feels we still need a human in the loop, then I am inclined to agree.

To which I replied:

Thanks for the recap Josh, I'd forgotten which call it was  😅

And if you want to defer to Andra, I don't feel strongly so Let's defer to @Andra.  As such, I'll follow her suggestions to require a translation before we merge.

Thanks all!

[andrablaj]

Just a small non-blocking suggestion. This is ready to go! Thanks for starting the conversation!