Needs #8 to be very useful.

Operations can be a lot faster if they're in place - because

1. Some vectorized operations are faster if they're in-place
2. No new array has to be allocated
3. No new NDArray object has to be allocated (objects are very heavy in Godot)

Since keyword arguments are not supported in gdscript right now (see #10), the best place to add in-place operations might be the object itself (with object.a syntax). This also makes it clear which object is the one affected by the operation. This is not behavior that is present in numpy and xtensor. But it might be useful for NumDot given its environment.