Fix assert in x86 when using custom align

[acoates-ms]

In x86 scenarios we have the Folly assert triggering in this code:

    FOLLY_ALWAYS_INLINE static uint64_t Read8(const uint64_t* buf, size_t off) {
      if constexpr (kHasUnalignedAccess) {
        uint64_t out;
        FOLLY_BUILTIN_MEMCPY(&out, buf + off, sizeof(out));
        return out;
      } else {
        assert(0 == reinterpret_cast<uintptr_t>(buf) % sizeof(*buf));
        return buf[off];
      }
    }

The analysis had shown that this assert is not correct for the x86 code where the allocated buffer can be aligned by 32-bits instead of 64-bits.

This PR overrides the SpookyHashV2.h so that we comment out the failing assert.
In practice it must not cause any issues:

• Intel CPUs allow unaligned memory access without issues. The assert prevents some operations that would normally work like in our cause with x86.
• ARM64 CPUs prohibit unaligned memory access, but if it happens, then we would have an immediate crash at this point. Thus, the assert is redundant.

Microsoft Reviewers: Open in CodeFlow