This shoud also fix issue #25: When the amount of data will be truncated
iconv() will return EINVAL when An incomplete multibyte sequence is
encountered in the input, and the input byte sequence terminates after
it. So if the input is larger than the internal buffer of Reader and
the end of the buffer conatins partial multi-byte chars, then Reader
will failed with EINVAL.
So when iconv() return EINVAL, we checks whether there are more data to
process, if so, we continue without report an error to user.
