perf: adds avx512 vector ops for koalabear and babybear fields

[gbotrel]

Description

The assembly is readable and a breeze to work with after doing the same things with multi words modulus.

Probably couple of perf aberration to correct and optimization to do.

Need to compare vec::mul with the awesome (and well documented :) ) work in Plonky3: https://github.com/Plonky3/Plonky3/blob/20256720b683897b634393dadcf8afab43101cb7/monty-31/src/x86_64_avx512/packing.rs#L319

Benchmark example

benchmark                                      old ns/op     new ns/op     delta
BenchmarkVectorOps/add_256-32                  141           14.7          -89.58%
BenchmarkVectorOps/sub_256-32                  282           14.8          -94.75%
BenchmarkVectorOps/scalarMul_256-32            299           62.8          -79.02%
BenchmarkVectorOps/sum_256-32                  224           30.5          -86.36%
BenchmarkVectorOps/innerProduct_256-32         496           74.9          -84.91%
BenchmarkVectorOps/mul_256-32                  299           66.8          -77.66%
BenchmarkVectorOps/add_512-32                  285           25.8          -90.96%
BenchmarkVectorOps/sub_512-32                  568           26.0          -95.43%
BenchmarkVectorOps/scalarMul_512-32            601           125           -79.23%
BenchmarkVectorOps/sum_512-32                  479           41.3          -91.39%
BenchmarkVectorOps/innerProduct_512-32         993           128           -87.12%
BenchmarkVectorOps/mul_512-32                  601           137           -77.22%
BenchmarkVectorOps/add_65536-32                39938         7742          -80.61%
BenchmarkVectorOps/sub_65536-32                80858         7730          -90.44%
BenchmarkVectorOps/scalarMul_65536-32          83196         16021         -80.74%
BenchmarkVectorOps/sum_65536-32                58330         2849          -95.12%
BenchmarkVectorOps/innerProduct_65536-32       133302        13518         -89.86%
BenchmarkVectorOps/mul_65536-32                86508         16781         -80.60%
BenchmarkVectorOps/add_524288-32               318606        68041         -78.64%
BenchmarkVectorOps/sub_524288-32               639760        68476         -89.30%
BenchmarkVectorOps/scalarMul_524288-32         664143        127940        -80.74%
BenchmarkVectorOps/sum_524288-32               464263        23079         -95.03%
BenchmarkVectorOps/innerProduct_524288-32      1068978       108061        -89.89%
BenchmarkVectorOps/mul_524288-32               689172        133001        -80.70%
BenchmarkVectorOps/add_1048576-32              638737        138234        -78.36%
BenchmarkVectorOps/sub_1048576-32              1282222       138241        -89.22%
BenchmarkVectorOps/scalarMul_1048576-32        1331820       256007        -80.78%
BenchmarkVectorOps/sum_1048576-32              924642        46221         -95.00%
BenchmarkVectorOps/innerProduct_1048576-32     2138163       215971        -89.90%
BenchmarkVectorOps/mul_1048576-32              1379577       266948        -80.65%
BenchmarkVectorOps/add_2097152-32              1298483       403597        -68.92%
BenchmarkVectorOps/sub_2097152-32              2599115       398351        -84.67%
BenchmarkVectorOps/scalarMul_2097152-32        2675315       514020        -80.79%
BenchmarkVectorOps/sum_2097152-32              1846965       92459         -94.99%
BenchmarkVectorOps/innerProduct_2097152-32     4279783       433708        -89.87%
BenchmarkVectorOps/mul_2097152-32              2801868       557975        -80.09%

[Tabaie]

If n is len(slice)/16 it should be documented.

[Tabaie]

Or better yet, just shift it by 4 in assembly instead of on the go side.