This PR consists of two parts that can be reviewed individually:

Decoupling the border size from the resize range

We place regions around the window in which the window borders are drawn. These regions currently stretch out as far as the resize range (e.g. how close to the window the pointer needs to be in order for it to be able to resize the window). We need to be able to draw shadows further than the resize range. This involves restricting the input region of the, now enlarged, border parts. Unfortunately, this is requires a breaking change in AdwaitaFrame::new as we need access to the wl_compositor for creating wl_region objects.

Generating and drawing the window shadow

I wanted the window shadows to be as close to libadwaita's window shadows as possible. But as far as I can see, its shadows are defined in (S)CSS, which is not an option for this crate. The solution to this I came up with is to measure libadwaita's shadow from a screenshot and use a curve-fitting algorithm to create an exponential function that lets us sample the alpha value of the shadow given the distance from the window's border.

To achieve this, I took these screenshots of a simple libadwaita window in both an active and inactive state (the shadow shrinks when the window is unfocused/inactive):

active.png	inactive.png

Next, I used this python script: https://gist.github.com/Friz64/ef941181f4d6aef82332f7066f984815

It's usage is as following:

./fit_libadwaita_shadow.py (active.png|inactive.png) [--plot]

When running it against active.png, it outputs the following:

a = 0.20650550475430265
b = 0.10461752988320006
c = -0.0005424462332409538
shadow size = 43

a, b, and c are the parameters for the exponential decay function a * exp(-b * x) + c. shadow size is the pixel distance from the window border at which the resulting u8 alpha value would round down to 0, making it meaningless to sample the function beyond this point.

Armed with this function, we now have the ability to draw the window shadow with any given scale factor. To do this, we generate the following two pixmaps for each encountered pair of scale factor and active state by sampling it at each pixel.

pixmap name / what it's used for	description	visual explanation
side	A strip of pixels, going from left to right with an increasing distance value.
edges	A square image, with the function sampled using the euclidean distance from the center. We offset this distance by the corner radius, so that we can place the edge shadow around the rounded window corners.

Then, all that's left is to copy these images (or parts of them) to the appropriate parts of the border, and we're done. We cache our work as much as possible in order to save resources.

Fixes #4.