MBS-10651: Use correct context for AI calls

[PhMemmel]

The current code uses the wrong context when sending AI requests. The context of the question is being used right now which is for example the context of the question bank the question is living in.

However, submitting an attempt in a quiz you definitely want the context of the quiz to be the one being used in the LLM request (which is being used for permission checks in local_ai_manager for example), because that's actually the context you are operating in.

Steps for testing:

• Install block_ai_control
• Install local_ai_manager
• Select local_ai_manager as backend for qtype_aitext in the admin settings
• Create course A with a qbank containing a qtype_aitext question
• Create course B with a quiz activity and add the question from the question bank from course A
• Check the "Add AI control center" in the course settings of course B.
• Enable all AI functionalitites in the control center block.
• Start an attempt as student, submit the attempt. You should not receive an error anymore (because now the correct context is being checked).

Bonus testing steps:

• Whenever making a request check the database entries in local_ai_manager_request_log / context tables.
• When submitting an attempt from inside a quiz, the contextid column in local_ai_manager_request_log table should have the quiz course module context id
• When doing a preview of the question in the question edit form or from the dropdown in the question bank list view, the contextin in local_ai_manager_request_log should be the context of the mod_qbank instance
• When doing a full quiz preview as teacher, the contextid in local_ai_manager_request_log should belong to the quiz course module
• Maybe you can think of other use cases :)

[marcusgreen]

Thanks Phillip I will look at it after Wednesday, when I am one of the presenters at the Moodle Leeds events (and will be showing some ByCS things). I don't doubt your code, I just want to understand it before I merge.

[PhMemmel]

No rush and thanks in advance! Final version will be pushed soon.

[PhMemmel]

I added some more context to the description. I think you got the issue now. It's now also easily testable manually. Unit test coverage should be pretty decent as well :)

[marcusgreen]

Thanks Phillip, I now understand your block ai_control plugin which I would not have done otherwise. I did a test with the Core AI subsystem and it worked as expected.

[PhMemmel]

Thanks for reviewing, testing and merging!