When decoding UTF-16, don't assume that the buffer is in native endianness

when checking surrogates.