Simplify encoding detection and source file opening by starting off reading bytes, and only switching to decoding once we know which encoding to use.