This is another round of improvements for xarch HWIntrinsics containment. In summary it:

• Simplifies IsContainableHWIntrinsicOp to use tuple type for size calculation where possible.
• Fixes several more places where a load was improperly contained when the instruction memory size requirement was larger than the load.
• Enables containment in more places it would have been valid but was missed before.
• Enables EVEX embedded broadcast for 64-bit constant vectors on x86.

Diffs look good.

Notable Fixes

These were overreading before and are fixed now:

-       vmovups  xmm0, xmmword ptr [rbp-0x190]
-       vcmpps   ymm0, ymm0, ymmword ptr [rax], 0
+       vmovups  xmm0, xmmword ptr [rax]
+       vmovups  xmm1, xmmword ptr [rbp-0x190]
+       vcmpps   ymm0, ymm1, ymm0, 0

-       vmovups  xmm0, xmmword ptr [rbp-0x190]
-       vcmppd   ymm0, ymm0, ymmword ptr [rax], 0
+       vmovups  xmm0, xmmword ptr [rax]
+       vmovups  xmm1, xmmword ptr [rbp-0x190]
+       vcmppd   ymm0, ymm1, ymm0, 0

(actually, this was both an overread bug and an incorrect SIMD size bug -- both fixed now)

Similarly:

static unsafe Vector256<long> ShouldNotContainBroadcast(int* ptr)
{
    return Avx2.BroadcastScalarToVector256(Sse2.LoadScalarVector128(ptr).AsInt64());
}

-       vpbroadcastq ymm0, qword ptr [rdx]
+       vmovd        xmm0, dword ptr [rdx]
+       vpbroadcastq ymm0, ymm0

And:

static unsafe Vector128<double> ShouldNotOverReadCreateScalarUnsafe(float* ptr)
{
    return Vector128.CreateScalarUnsafe(Sse.LoadScalarVector128(ptr).AsDouble().ToScalar());
}

-       vmovsd   xmm0, qword ptr [rdx]
+       vmovss   xmm0, dword ptr [rdx]