x86: Add 8-bit ipred z3 AVX-512 (Ice Lake) asm

The z3 algorithm is column-based in nature, but by leveraging the AVX-512 permutation instructions we can convert it into a row-based implementation to avoid having to transpose the output. This results in a decent improvement over the existing AVX2 code:

intra_pred_z3_w4_8bpc_c:             208.9 ( 1.00x)
intra_pred_z3_w4_8bpc_ssse3:          53.0 ( 3.94x)
intra_pred_z3_w4_8bpc_avx2:           37.5 ( 5.57x)
intra_pred_z3_w4_8bpc_avx512icl:      25.9 ( 8.07x)

intra_pred_z3_w8_8bpc_c:             609.3 ( 1.00x)
intra_pred_z3_w8_8bpc_ssse3:          81.5 ( 7.48x)
intra_pred_z3_w8_8bpc_avx2:           59.8 (10.19x)
intra_pred_z3_w8_8bpc_avx512icl:      32.1 (19.01x)

intra_pred_z3_w16_8bpc_c:           1644.7 ( 1.00x)
intra_pred_z3_w16_8bpc_ssse3:        172.5 ( 9.54x)
intra_pred_z3_w16_8bpc_avx2:         106.4 (15.46x)
intra_pred_z3_w16_8bpc_avx512icl:     54.2 (30.33x)

intra_pred_z3_w32_8bpc_c:           3569.7 ( 1.00x)
intra_pred_z3_w32_8bpc_ssse3:        351.4 (10.16x)
intra_pred_z3_w32_8bpc_avx2:         210.7 (16.94x)
intra_pred_z3_w32_8bpc_avx512icl:     99.9 (35.74x)

intra_pred_z3_w64_8bpc_c:           8474.9 ( 1.00x)
intra_pred_z3_w64_8bpc_ssse3:        768.6 (11.03x)
intra_pred_z3_w64_8bpc_avx2:         451.1 (18.79x)
intra_pred_z3_w64_8bpc_avx512icl:    203.5 (41.64x)