Unlike other variants, for YUY2 we need to use different prediction:
* on line 0 for luma we should left predict starting from the second pixel
* on line 1 we should left predict first 4 pixels for luma and 2 for chroma
* median prediction employed here is taken directly from HuffYUV