Thanks I was wondering will this

Yes, it will!

In normal list-mode, it will check every piece of data. This is usually good for summaries.

The embedding mode will only check the closest matching text chunks. By default this is 1, but you can set it in the query with something like similarity_top_k=2

My concern here is that i will provide the LLM a context that is not larger than 4096 tokens and making a single call. There are probably over 100k token on my data and it will only extract topics of the "sample" given in the context

Correct! llama index handles this by splitting your text into "chunks" (with some overlap) so that the token limit is never reached

Then, it uses a refine procedure to send multiple chunks and refine a response to a query

I recommend checking out this page: https://gpt-index.readthedocs.io/en/latest/guides/index_guide.html

And at the bottom, response synthesis is covered: https://gpt-index.readthedocs.io/en/latest/guides/index_guide.html#response-synthesis