feat: Enable auto translation on learner facing content

[zamanafzal]

What are the relevant tickets?

https://github.com/mitodl/hq/issues/9667

Description (What does it do?)

This PR adds automated translation management functionality to the ol_openedx_course_translations plugin. The system syncs translation keys from English to target language files, translates empty keys using LLM APIs, and automatically creates pull requests in the mitxonline-translations repository.

Relevant Test Pull Requests:
https://github.com/zamanafzal/mitxonline-translations/pulls

How can this be tested?

Ensure you're in the Open edX environment (Tutor container)

Checkout this branch for open-edx-plugins.
tutor dev exec lms bash
export the LLM key you want to use. e.g MISTRAL
export MISTRAL_API_KEY="KEY"
Generate a github token which will be used to authenticate with github and pushing changes.
export GITHUB_TOKEN ="token"
./manage.py sync_and_translate_language {lan_name} --model {model_name}

For example:

 ./manage.py lms sync_and_translate_language ar --model gemini/gemini-3-pro-preview
./manage.py lms sync_and_translate_language ar  --model mistral/mistral-small-latest

It will translate the missing keys and generate files if don't exist and create a PR.

[asadali145]

As discussed over the call, Mistral needs some changes on how default model is set. It is not working when we remove mistral/ from the model name. Also, gemini is not working.

[asadali145]

LGTM!

[asadali145]

Why are we including vertex_ai here?

[zamanafzal]

LiteLLM allows access to Gemini in two ways:

Gemini API (prefix: gemini/)

Vertex AI (prefix: vertex_ai/)

For example, the model vertex_ai/gemini-3-pro-preview uses Vertex AI. If only gemini-3-pro-preview is provided, LiteLLM may default to using Vertex AI.

[gemini-code-assist]

The author's explanation is correct. LiteLLM, the library used here, supports accessing Gemini models through both the native Gemini API (prefixed with gemini/) and Google Cloud's Vertex AI (prefixed with vertex_ai/). The code ensures that if a Gemini model is specified without either of these prefixes, it defaults to the gemini/ prefix for consistency with LiteLLM's expected format, while still allowing explicitly specified vertex_ai/ models to pass through unchanged.

[asadali145]

Can we explain the issues here?