riscv64/mc: Add bidir functions

This code compromises between the performance of a dedicated kernel per VLEN/width pair, and the flexibility of a fully VLEN-dynamic loop, by using a special case for w=4, and subdividing the rest into the w8/16/32 fast-paths (unrolled four lines per iter) and the w64/128 slow paths (one line per iter). For vl256, the fast path also covers w64.

Kendryte K230

avg_w4_8bpc_c:        346.7 ( 1.00x)
avg_w4_8bpc_rvv:       56.3 ( 6.16x)
avg_w8_8bpc_c:       1054.7 ( 1.00x)
avg_w8_8bpc_rvv:      139.0 ( 7.59x)
avg_w16_8bpc_c:      3398.6 ( 1.00x)
avg_w16_8bpc_rvv:     350.1 ( 9.71x)
avg_w32_8bpc_c:     13726.7 ( 1.00x)
avg_w32_8bpc_rvv:    1246.9 (11.01x)
avg_w64_8bpc_c:     33217.0 ( 1.00x)
avg_w64_8bpc_rvv:    3789.6 ( 8.77x)
avg_w128_8bpc_c:    83483.2 ( 1.00x)
avg_w128_8bpc_rvv:   9786.0 ( 8.53x)

w_avg_w4_8bpc_c:        441.6 ( 1.00x)
w_avg_w4_8bpc_rvv:       72.5 ( 6.10x)
w_avg_w8_8bpc_c:       1364.8 ( 1.00x)
w_avg_w8_8bpc_rvv:      200.3 ( 6.81x)
w_avg_w16_8bpc_c:      4417.6 ( 1.00x)
w_avg_w16_8bpc_rvv:     562.7 ( 7.85x)
w_avg_w32_8bpc_c:     17890.7 ( 1.00x)
w_avg_w32_8bpc_rvv:    2093.5 ( 8.55x)
w_avg_w64_8bpc_c:     43231.4 ( 1.00x)
w_avg_w64_8bpc_rvv:    5739.2 ( 7.53x)
w_avg_w128_8bpc_c:   107984.3 ( 1.00x)
w_avg_w128_8bpc_rvv:  14283.1 ( 7.56x)

mask_w4_8bpc_c:        497.5 ( 1.00x)
mask_w4_8bpc_rvv:       92.8 ( 5.36x)
mask_w8_8bpc_c:       1529.7 ( 1.00x)
mask_w8_8bpc_rvv:      253.4 ( 6.04x)
mask_w16_8bpc_c:      4964.3 ( 1.00x)
mask_w16_8bpc_rvv:     680.4 ( 7.30x)
mask_w32_8bpc_c:     20288.7 ( 1.00x)
mask_w32_8bpc_rvv:    3015.6 ( 6.73x)
mask_w64_8bpc_c:     49726.6 ( 1.00x)
mask_w64_8bpc_rvv:    7210.9 ( 6.90x)
mask_w128_8bpc_c:   126687.1 ( 1.00x)
mask_w128_8bpc_rvv:  18279.9 ( 6.93x)