It is obvious to me: I think they chose to load, parse, prettify, colour etc. and keep the entire file into an in-memory data structure, which is easy to reason about, and totally reasonable for "human sized" data. However, for arbitrary sized files, you need a mechanism for leaving most of the file in a raw state - that's harder (but not impossible, obviously).

So, it makes sense to me that they would assume only "human sized" data (and expect users to use another tool for looking at large files, "do one thing well" and all that) and opt for a slightly simpler programming model. Apparently, they decided that the assumption was wrong, and so changed the model.