Why, if the dimension of matrices is 3x3 for using a translate matrix, the correct result is obtained when we multiply a vector by a matrix, and when using 4x4 matrices, the correct result is obtained when we multiply a matrix by a vector?

```
func check() {
let delta: SIMD2<Float> = [100, 100]
var matrix3x3 = matrix_identity_float3x3
matrix3x3[0, 2] = delta.x
matrix3x3[1, 2] = delta.y
var matrix4x4 = matrix_identity_float4x4
matrix4x4[3, 0] = delta.x
matrix4x4[3, 1] = delta.y
let dote3: SIMD3<Float> = [10, 10, 1]
let dote4: SIMD4<Float> = [10, 10, 0, 1]
let r1 = matrix3x3 * dote3 // SIMD3<Float>(10.0, 10.0, 2001.0) // bad
let r2 = dote3 * matrix3x3 // SIMD3<Float>(110.0, 110.0, 1.0) // good
let r3 = matrix4x4 * dote4 // SIMD4<Float>(110.0, 110.0, 0.0, 1.0) // good
let r4 = dote4 * matrix4x4 // SIMD4<Float>(10.0, 10.0, 0.0, 2001.0) // bad
}
```