poly ntt invntt scheduling

[tfaoliveira]

this was done on top of the keccak branch (it should have been done differently).

The cycles improvements are the following:

keypair       , 220384 -> 213394
enc           , 267904 -> 261994
dec           , 339528 -> 330058

which help in closing the gap between the comparable C implementation from pq-crystals/kyber/standard (compiled with -mno-sse and bmi is also disabled):

keypair, 210982
enc    , 252650
dec    , 326310

[tfaoliveira]

this was done on top of the keccak branch (it should have been done differently).

I can try to fix this on Sunday.

[tfaoliveira]

thanks a lot!