Reorder some elements in the x264_mb_analysis_list_t struct to reduce the
amount of padding required.

Also drop the MMX implementation in favor of SSE.