Refactored reading of single elements from a connection.

Now, everything is again byte-based (even in a text-mode connection) and the read bytes are then encoded at once.
This makes the code much cleaner.
However, this assumes that a multi-byte character cannot contain any of the byte signatures of NUL '0x00' or line feed '0x0A'.
But I haven't seen this in the UTF-8/UTF-16 tables.