Better character set encoding detection

Id3 tags are supposed to be ISO-8859-1 or unicode, but often aren't.
To better detect the real encoding we now use ICU to detect possible
encodings for a given byte sequence, then apply additional heuristics
to determine the most likely one.
b/5564857

Change-Id: I53bc83b006433da5c2f2ccfcd770ddb3a26b64d0