twofish-avx2: de-unroll round function

* cipher/twofish-avx2-amd64.S (__twofish_enc_blk16)
(__twofish_dec_blk16): Use loop structure instead of unrolling.

De-unrolling reduces code-size significantly and gives
small (<1%) increase in speed (tested on zen4, tiger-lake).

• Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>