This isn't something a toolkit could implement itself.
Imagine a compositing window manager that takes a particular window and renders it at twice its original dimensions. Input redirection is needed to let the user interact with this zoomed window in a way that the underlying application can understand (halve the x and y coordinates and translate to match the real window position).
All this has to happen before the input reaches the application. After all, we want clicks that occur outside the application's window to be directed to it.