Move to the new gfx-rs swapchain API

[kvark]

See https://gfx-rs.github.io/2019/10/01/update.html

Quad example update - https://github.com/gfx-rs/gfx/commit/40a905ff0ccd583f18ced4250c27b7765716a16f#diff-3396f7c6f19ad0882c9d181e08c8c702

[zakorgy]

With the new swapchain API we can't directly access the frame images. Steps needs to be done before we can move to the new API:

• [x] Replace pipeline barriers on frame images with render passes.
• [x] Make a workaround for blitting from/to frame images. (using shaders?)
• [x] Rework the read_pixel logic which we use in wrench, because we can't directly read the frame image content into a buffer. Instead we render to a different image which we can read back and copy the result from that image to the screen.

[kvark]

You are right that we can't have this new swapchain model for as long as we need to blit back from the screen :/ And I think it's only the MixBlend mode (with Readback task) that actually needs to blit from the screen. There are two ways to solve this that I could see:

1. reword the new swapchain model to allow more usages and expose images... this means we'll have to request the ability to read images back in WR unconditionally, which potentially comes at a cost.
2. in WR force the surface allocated if we see a MixBlend mode. That means that any readbacks would only need to be done from a render target.

[kvark]

More stuff to read about (1) - https://github.com/gpuweb/gpuweb/issues/437

[zakorgy]

Fixed in #331