Yes, prompts may contain formulas and formulas can carve out data and content from document sources, of course.
Yes.
Yes. I believe it is about 4k (characters). Codans need to chime in with the actual size. I have attempted to use large data sets (> 4k) - no go.
Some day, but not today. This article touches on your dreams by using aggregations of large data sets to make inferences from.