Update glfw/vulkan backend to use explicit context and add single-threaded multiple context demo.

[FunMiles]

In this PR, the first commit adds the backend support in glfw/vulkan for multiple contexts to the docking branch.
The second commit changes the demo app to show a running example of three contexts each running a demo-window with docking enabled.

Note that the second diff creates a new class ThreadWindow. This class will be used almost unmodified in an upcoming demo with multi-threading active. The difference is that in this commit, wait_for_main which runs a functor on the main thread executes directly, as there is only one thread. In the multi-threading demo, that function passes the functor to the main thread and waits until it has finished.

[FunMiles]

The third commit demoes multi-threaded Vulkan work combined with imgui.
Note that the approach of running on the main thread all interactions with the event system and for presenting the graphics impacts the frame rate at least on Mac-OS. Compare:

• with one context/producer thread, 120fps
• with two contexts/producer threads, 60fps
• with three contexts/producer threads, 40fps

[FunMiles]

The performance is improved with having a shorter lock on the mutex for main thread job queue.