Improve AddImpl and SubtractImpl with Avx512

[benaadams]

Removes three instructions from the filling out of the instruction set with Avx512

; Method Nethermind.Int256.UInt256:AddImpl(byref,byref,byref):ubyte (FullOpts)
G_M000_IG01:                ;; offset=0x0000

G_M000_IG02:                ;; offset=0x0000
       vmovups  ymm0, ymmword ptr [rcx]
       vpaddq   ymm1, ymm0, ymmword ptr [rdx]
-      vpternlogq ymm3, ymm3, ymm2, 85
-      vmovaps  ymm4, ymm0
-      vpternlogq ymm4, ymm3, ymm1, -56
-      vpternlogq ymm0, ymm4, ymm1, -20
+      vpcmpuq  k1, ymm1, ymm0, 1
+      vpmovm2q ymm0, k1
       vmovmskpd rax, ymm0
       vpcmpeqd ymm0, ymm0, ymm0
       vpcmpeqq ymm0, ymm0, ymm2
       vmovmskpd rcx, ymm0

       vpcmpuq  k1, ymm1, ymm0, 1
       vpmovm2q ymm0, k1
       vmovmskpd rax, ymm0
       vpcmpeqd ymm0, ymm0, ymm0
       vpcmpeqq ymm0, ymm0, ymm1
       vmovmskpd rcx, ymm0

       lea      eax, [rcx+2*rax]
       xor      ecx, eax
       and      ecx, 15
       movsxd   rcx, ecx
       shl      rcx, 5
       mov      rdx, 0x2272EAEC928
       vmovups  ymm0, ymmword ptr [rcx+rdx]
       vpaddq   ymm0, ymm2, ymm0
       vmovups  ymmword ptr [r8], ymm0
       test     al, 16
       setne    al
       movzx    rax, al

G_M000_IG03:                ;; offset=0x0065
       vzeroupper 
       ret      
; Total bytes of code: 105

[rubo]

Have you tested on supported hardware?

[benaadams]

Have you tested on supported hardware?

Yes, is supported by AMD Ryzen Zen 4+