ARM64: bad codegen for Vector.Dot

The following code returns a different value under jit optimizations on arm64:
```
public static int Main()
{
int result = Vector128.IndexOf(
AdvSimd.SubtractRoundedHighNarrowingUpper(Vector64.AllBitsSet,
Vector128.AllBitsSet,
Vector128.AllBitsSet),
Vector.Dot(Vector.One, Vector.One));

return result == -1 ? 100 : 101;
}
```

jit generates CMEQ for Vector IndexOf which expects the dot product in every lane.
Lowering for this Vector.Dot uses AddAcross (ADDV) which writes to only lane 0.
Either we shouldn't have dropped the toScalar+Create during Morph, or lower must explicitly broadcast as well.