x86: Improve high bitdepth cdef_filter AVX2 asm

                              old  ->  new
cdef_filter_4x4_16bpc_c:     779.5
cdef_filter_4x4_16bpc_ssse3:  94.4
cdef_filter_4x4_16bpc_avx2:  107.1     73.3
cdef_filter_4x8_16bpc_c:    1450.0
cdef_filter_4x8_16bpc_ssse3: 157.9
cdef_filter_4x8_16bpc_avx2:    n/a    107.2
cdef_filter_8x8_16bpc_c:     581.4
cdef_filter_8x8_16bpc_ssse3: 261.3
cdef_filter_8x8_16bpc_avx2:  252.1    159.9