Remove unsound Send & Sync impls

[jplatte]

The unsound impls weren't really being relied upon because it was ensured that WindowState was never truly moved or accessed across thread boundaries, but this PR changes things such that they're no longer required, preventing the impls from causing issues in the future.

Fixes #2067.

[jplatte]

I made sure that the hello example still starts up successfully on Linux. Should I test anything else to make sure the panic is not hit?

[jplatte]

Oh, seems like I broke something again. Will look into it in a bit.

[maan2003]

I made sure that the hello example still starts up successfully on Linux. Should I test anything else to make sure the panic is not hit?

try timer, sub window, slow function blocking function and async_event examples

[jplatte]

Okay, seems like this approach doesn't work because the idea is really that any thread can register a callback with IdleHandle::add_idle_callback, and that callback will end up being called on the UI thread. I think the unsafe impls could potentially be removed by splitting WindowState into a Send part and a non-Send part, I'll try that in the coming days.

[jneem]

I'm a bit confused by this clippy lint, to be honest. Isn't the whole point of unsafe impl Send to say "I know that some of the fields are !Send, but I promise to use them safely"? Is there any usage of unsafe impl Send that doesn't trigger the lint?

Edit: I found some discussion here. Indeed, it seems like the lint will be demoted until some issues with false positives are resolved.

[jplatte]

Makes sense. I was also a bit confused, but at the same time I thought that this impl seems rather dangerous. I'll close this for now since I won't have time to try the alternative approach I mentioned above in the near future.