Make `Window` and `Windows` structs easier to test with

[alice-i-cecile]

Writing tests for these types is quite challenging.

We should have:

1. Constant constructors wherever possible.
2. Doc tests demonstrating simple testing practices (likely using the Default struct update syntax for WindowDescriptor.
3. A const fn new method on Windows that accepts a single Window.

[alice-i-cecile]

The sticking point is the RawWindowHandle type: we cannot reasonably construct this in tests. IMO this should be optional in order to enable headless testing of various utility methods.

[targrub]

I'm taking a look at tackling this.