This prevents WMs like KWin or Mutter from automatically maximizing windows which frames are larger than their monitor area.
The maximization otherwise gets fed back into the win32 state, and often confuses applications which are operating a fullscreen window change by resizing a window then later changing its decoration.
It isn't possible to tell between a maximization initiated by the user and one that is initiated by the WM responding to a window resize from us. This heuristic should hopefully be good enough, as it is unlikely that the user will ever be able to resize a window in a such a way that its *frame* covers a monitor entirely. Note that if the user resizes a window such that its *visible* area covers a monitor entirely, we are going to make it fullscreen already anyway.