Handle Text generation for Lecture Slides and HWs differently

[XThomasBU]

• Add information in the metadata for each chunk mentioning what type of content (Lecture Slide, HW, logistics, etc) it holds
• Based on this the LLM text generation is handled differently, by specifying how to handle different context types in the prompt

[Farid-Karimli]

This is particularly important for lecture slides, since some slides are mostly blank and build up to the final slide content (think heading, then list items, then the whole thing). Would need some kind of duplicate processing/filtering.

[XThomasBU]

One possible solution, from the index retrieve the content type (from the metadata), and build the response strategy based on the content type. Superficial answers if the content type == Assignment, vs in-depth answers if content type == lecture.