Upgrade to bevy 0.10

[kulkalkul]

This is what I've done as a follow-up on #4

So, I've upgraded iced's wgpu in a fork:
https://github.com/kulkalkul/iced/commits/self-usage
I've done it, so it is non-breaking. Those expect calls there are similar in how it handles pre-wgpu 0.14. It is also handled the same in
https://github.com/sconybeare/wgpu_glyph/tree/wgpu-0.15

For bevy_iced, there are two changes I think you might be interested in:

• I've changed Res<Windows> to Query<Window> and kept it a single window (I have no idea how I can do it for multiple windows).
• Bevy decided to hide RenderContext fields, so they can only be accessed with functions ([Merged by Bors] - Support recording multiple CommandBuffers in RenderContext bevyengine/bevy#7248). But the command_encoder function is &mut self, so it isn't possible to get both device and command_encoder from RenderContext. So, I've got the RenderDevice from the world, which is OK but feels a little hacky tbh.

Aside from the above points, a bug causes flashing rendering for UI. It happens randomly, at a random rate, sometimes so slow it is unnoticeable. Here is the video of it:

FAST FLICKERING VIDEO, EPILEPSY/SEIZURE WARNING!

magellan_2023-03-23_11-28-49.mp4

---

I've figured that the new pipelined rendering implementation causes IcedNode::run to run parallel to IcedContext::display. As a result, the run function misses the IcedProp.did_draw change. I didn't know the project enough to solve it, so I decided to ask it.

I hope this isn't too long 😅

[kulkalkul]

Each line is frame boundary. T(rue) and F(false) represents last state of did_draw. So T happens when IcedContext::display runs, F happens when IcedNode::run runs:

I thought about this for a while. I can only think of creating a did_draw state with three possible values. I've thought about finding another way to do this, as this approach delays all did_draw changes by one frame. But that's only possible to detect by knowing which order they will execute, which is impossible with new multithreaded rendering.

did_draw depends on user input set to true, but this is not the case for setting it to false. In theory, the user can state it, but in practice, having a function named "no_display" doesn't make sense. So, having a meaningful API to control this behaviour is impossible, so I accepted that a frame delay is the only fix.

[tasgon]

Thank you so much for the contribution. I was also a bit worried about the synchronization, so I tried a different trick. If you're interested, take a look at #6 and let me know which method you prefer.

Also, I'm fine with using your iced fork for now, but I will ask for a specific ref to pin those crates to. Is the current head on that branch okay?

[kulkalkul]

I don't know how extracts are synchronized, but it works! 🥳 I prefer your version, as that's probably how this should've been.

I didn't do anything after that commit, so the current head is okay. I should probably open a PR to iced, but the change is so minimal I'm not sure if it is worth the hassle.

[tasgon]

Closing as #6 has been merged.