Remove RA288Context.output buffer. This buffer is just RA288Context.sb backwards (output[i] == sb[N-i], where N is the buffer length). This makes the code slower, this will be fixed in my next commit.